Imported from marktantongco/aiskills-photog (SKILL.md). Install upstream with npx skills add marktantongco/aiskills-photog. Copyright stays with the author.

AI Practitioner Skills Framework

A comprehensive, structured reference for core AI skills, sub-skills, and competencies. Designed for learners, professionals, and AI agents.

Last updated: August 2026 | Version: 5.2 | Format: SKILL.md | 7 Domains · 76+ Sub-skills

Philosophy of Professional Visual Engineering

The shift from "AI Artist" to "Professional Visual Engineer" represents a fundamental reorientation: the practitioner treats AI generation not as magic, but as a precision instrument that responds to physics, psychology, and deliberate architectural choices.

Core Principles:

Photorealistic Logic First: Every generation operates within the laws of optics, anatomy, and spatial psychology. The AI models physics; you must understand the physics the AI is modeling.
Quiet Luxury of Execution: Professional work reveals itself through restraint -- precise control rather than maximum effect. The mark of expertise is knowing what not to prompt.
Intentional Architecture: Structure prompts as you'd architect a building -- load-bearing concepts first, decorative elements last. Token order isn't cosmetic; it's structural.
Verification as Discipline: Generation is the beginning, not the end. Systematic quality gates, iteration loops, and mastery checklists separate professionals from enthusiasts.

This framework treats AI image generation as a professional engineering discipline: systematic, physics-informed, quality-gated, and continuously improvable.

Quick Reference: Skill Synergy Map

+------------------------------------------------------------+
|                       FOUNDATION                            |
|  Technical Prompt Engineering + Photographic Literacy       |
+-------------------+--------------------+-------------------+
                    |                    |
                    v                    v
+---------------------------+ +---------------------------+
|     CONSISTENCY LAYER     | |     REFINEMENT LAYER      |
|    Strategic Negation     | |     Post-Processing +     |
|  + Identity Preservation  | |     Hybrid Workflows      |
+-------------+--------------+ +-------------+--------------+
              |                              |
              v                              v
+------------------------------------------------------------+
|                  ORCHESTRATION LAYER                        |
|        AI Agent Design + Production Deployment              |
+------------------------------------------------------------+

Key Insight: Skills compound. Mastery of foundational prompting enables effective consistency control; consistency enables reliable hybrid workflows; reliable workflows enable scalable agent orchestration.

Core Skill Categories

1. Technical Prompt Engineering

Constructing prompts as structured blueprints, not keyword lists. The input layer for all AI generation.

Sub-Skill	Explanation	Tools & Techniques	Example
Scaffold Method	Structured prompt order: `[Subject] + [Action] + [Lighting] + [Lens/Specs] + [Style] + [Quality]`	Prompt templates, style guides, token-aware editors	`A woman walking in rain + cinematic lighting + 85mm f/1.8 + photorealistic + 4K`
Information Priority (Front-Loading)	Place critical elements first; AI weights early tokens more heavily	Token analysis, prompt testing	`Oil painting of a castle...` (style first) vs. `A castle, oil painting...`
Photographic Vocabulary	Use precise technical terms over vague buzzwords	Lens/aperture glossaries, photography references	`85mm lens, f/2.8, shallow depth of field` instead of `hyperrealistic`
Active Voice Editing	Use clear action verbs for iterative edits and refinements	Inpainting prompts, edit command syntax	`remove the background`, `add a red hat`, `turn the car into a truck`
Agent Prompting Patterns	Design prompts for multi-step agent workflows using structured reasoning	Zero-shot, few-shot, chain-of-thought, role prompting	`You are a cinematographer. First analyze the scene mood, then specify lighting.`
Long-Horizon Video Scripts	Maintain narrative coherence across extended video generation sequences	Dialogue-to-script translation, scene continuity prompts, beat sheets	`Scene 1: Close-up on protagonist, golden hour. Transition: Cross-dissolve to wide establishing shot.`
Camera Movement Language	Specify complex cinematography techniques in text prompts	Camera direction lexicon, film terminology guides	`pull from close-up to wide shot`, `circular tracking shot`, `dolly zoom`
Advanced Prompt Syntax	Master weighted syntax, conditional prompts, and parameter chaining for precise control	Weight syntax `::`, conditional branches, parameter interpolation, prompt variables	`(masterpiece:1.5), (best quality:1.3), [cinematic lighting] :: 0.8`
Semantic Layering	Stack semantic concepts in deliberate order to control interpretation hierarchy	Concept ordering, modifier placement, implicit vs explicit directives	`A [serious executive] [in shadows] [dramatic side lighting]` creates different emphasis than `[dramatic side lighting] [serious executive] [in shadows]`
Style Injection Techniques	Embed artistic references without overriding subject integrity	Style-strength anchors, reference blending, preservation weights	`::0.3` after style reference to preserve subject while adding aesthetic influence

Connects to: Photographic Literacy (shared vocabulary), Identity Preservation (consistent character direction), Agent Orchestration (scripted agent tasks)

2. Photographic Literacy

Reconstructing real-world physics to achieve believable, professional-grade results.

Sub-Skill	Explanation	Tools & Techniques	Example
Lighting Pattern Mastery	Apply classic studio lighting setups to sculpt form, mood, and dimension	Rembrandt, Butterfly, Rim, Split, Loop lighting; 3-point lighting diagrams	`Rembrandt lighting: key light 45deg high, fill at 1/4 power, subtle rim for separation`
Lens Selection	Choose focal length for desired perspective, compression, and emotional effect	Portrait (85-135mm), Standard (35-50mm), Wide (18-24mm); lens simulators	`85mm portrait lens for flattering facial compression and background bokeh`
Focal Length Mapping	Understand how focal length affects spatial relationships, depth perception, and emotional tone across different subject distances	Compression ratios, perspective distortion curves, subject-to-camera distance mapping	24mm at 2m = dramatic wide-angle distortion; 135mm at 5m = flattering compression with natural background separation
Optical Physics for AI	Apply real lens characteristics -- bokeh quality, sphere vs. anamorphic rendering, diffraction limits -- to prompt believable results	Lens render characteristics, circle of confusion, MTF charts, anamorphic streak patterns	Prompt: `shot on Leica 50mm Summilux, visible anamorphic horizontal flares, organic bokeh circles`
Aperture Control	Use f-stop to control depth of field and subject isolation	Shallow (f/1.4-f/5.6), Deep (f/8-f/32); DOF calculators	`f/2.8 for subject isolation; f/11 for product shots requiring full sharpness`
Native Resolution Management	Select output resolution based on use case; understand native detail rendering	720p (preview), 1080p (enhanced), 4K (professional); resolution tier selection	`Generate at 4K native to capture skin pores and fabric weave; avoid upscaling artifacts`
Vertical Format Mastery	Compose specifically for 9:16 short-form platforms with platform-aware framing	Subject placement strategies, negative space planning, vertical leading lines	`Center subject with negative space above for TikTok text overlays; use leading lines vertically`
Advanced Rendering Terms	Use physics-based cues for realistic material response and light interaction	Global illumination, ray tracing, ambient occlusion, subsurface scattering	`subsurface scattering on skin/ears; ambient occlusion in fabric folds and crevices`
Anamorphic Mastery	Apply cinematic widescreen lens characteristics -- horizontal flares, elliptical bokeh, distinctive lens breath -- for dramatic scope and visual signature	Anamorphic lens simulation, flare placement, aspect ratio conventions (2.39:1, 2.76:1), squeeze factor awareness	Prompt: `shot on Panavision Primo anamorphic, horizontal blue lens flares, 2.39:1 aspect ratio, oval bokeh highlights`
Cinematic Grammar	Use film language conventions -- coverage, shot progression, visual rhythm -- to construct narrative meaning across sequences	Shot-reverse-shot, 180-degree rule, match cuts, visual motifs, color scripting	`Open on wide establishing shot, push in to medium, cut to close-up reaction; parallel editing across two locations`
Executive Portraiture	Compose authority portraits using psychological signaling -- power poses, deliberate lighting hierarchy, environmental context cues	Status signaling, environmental framing, wardrobe psychology, gaze direction, spatial dominance	Prompt: `CEO portrait, corner lighting establishing authority, dark navy suit against library background, slight upward camera angle, confident gaze at viewer`
Quiet Luxury Aesthetic	Execute understated wealth signaling through restraint -- muted palettes, natural materials, absence of logos, tactile quality emphasis	Minimal branding, heritage fabrics, neutral tones, intentional simplicity, craft appreciation	Prompt: `quiet luxury editorial, cream cashmere sweater, matte gold jewelry, no visible logos, soft north window light, textured linen backdrop`

Connects to: Technical Prompt Engineering (shared terminology), Strategic Negation (realism standards), Post-Processing (foundation for edits)

3. Strategic Negation & Skin Mastery

Telling the AI what not to include to overcome unnatural perfection and maintain visual integrity.

Sub-Skill	Explanation	Tools & Techniques	Example
Negative Prompting with Weights	Actively exclude unwanted elements using weighted syntax for precision control	`(unwanted:weight)`, platform-specific negative prompt fields, weight tuning	`(plastic skin:1.4), (cartoon:1.3), (extra fingers:1.4)`
Skin Realism Management	Simultaneously prompt for natural textures while negating synthetic artifacts	Texture keyword pairs, reference anchoring, skin study libraries	Prompt: `visible pores, fine vellus hair, subtle skin variation` Negate: `airbrushed, doll-like, waxy, plastic`
Anatomical Correction	Proactively target common AI generation errors before they appear	Error-specific negative terms, anatomical reference sheets, validation checks	`(fused fingers, extra limbs, double iris, asymmetrical eyes:1.4)`
Temporal Consistency for Video	Prevent feature drift across video frames using stabilization techniques	Reference locking, error recycling, frame interpolation, motion vectors	Use `--cref` with character sheet + `retraining by error recycling` for stable sequences
Drift Correction	Feed generation artifacts back into models to improve long-sequence stability	Iterative refinement loops, frame-by-frame feedback, artifact analysis	Generate 10-frame clip, identify drift points, retrain prompt weights, regenerate with corrections

Connects to: Identity Preservation (consistency foundation), Photographic Literacy (realism standards), Agent Orchestration (automated quality control)

4. Identity Preservation & Consistency

Maintaining specific characters, styles, and narratives across multiple generations and scenes.

Sub-Skill	Explanation	Tools & Techniques	Example
Seed Locking	Fix initial noise pattern for controlled variations while changing other parameters	`--seed` parameter, deterministic generation, seed databases	`--seed 12345` to reproduce base composition while varying style modifiers or lighting
Reference Tools	Use platform-specific reference systems for identity and style preservation	Midjourney `--cref`/`--sref`, SD ControlNet/IP-Adapter, reference embedding	`--cref character_sheet.jpg --sref art_style.jpg --cw 80`
Character Weight (`--cw`)	Fine-tune reference influence scope from face-only to full appearance preservation	`--cw 0` (face only) to `--cw 100` (full appearance); weight testing	`--cw 80` to preserve face and clothing but allow pose variation and expression changes
Multi-Reference Consistency	Combine multiple references for complex scene coherence across characters, objects, locations	Up to 4 reference images per generation; reference blending strategies	`char1.jpg + char2.jpg + background.jpg + prop.jpg` for ensemble scene with consistent elements
Agent-Based Consistency	Leverage specialized agents for narrative coherence across complex multi-scene projects	ScripterAgent, DirectorAgent, Visual-Script Alignment (VSA), coordination protocols	DirectorAgent coordinates Veo 3.1 + Midjourney for cross-scene visual continuity and script fidelity
Visual-Script Alignment (VSA)	Evaluate and maintain faithfulness between generated visuals and source scripts/storyboards	Similarity metrics, human-in-the-loop checks, automated scoring, deviation thresholds	Score generated frames against storyboard; flag deviations >15% for review and regeneration

Connects to: Strategic Negation (drift prevention), Post-Processing (consistency refinement), Agent Orchestration (multi-agent narrative control)

5. Post-Processing & Hybrid Workflows

Treating AI generation as the start of an iterative, multi-tool creative process -- not the final step.

Sub-Skill	Explanation	Tools & Techniques	Example
Iterative Refinement (Branching)	Make small incremental changes while preserving core composition and seed	Seed locking, prompt diffing, version control, A/B testing workflows	Keep `--seed 12345`; change `weather: rainy` to `snowy`; compare outputs side-by-side
Inpainting & Outpainting	Perform surgical edits to fix errors or expand canvas beyond original generation bounds	Midjourney Vary Region, Photoshop Generative Fill, SD Inpaint, masked editing	Fix malformed hands via targeted inpainting; extend background via intelligent outpainting
External AI Enhancement	Use specialized tools for final polish, upscaling, and professional finishing	Topaz Photo AI (upscaling/denoising), Lightroom (grading), Runway (motion), ACE++	Upscale 4K native gen to 8K with Topaz; color grade in Lightroom for cinematic LUT application
Unified Generation-Editing	Generate, edit, and refine within same model context to reduce tool-switching friction	Qwen Image 2.0, Gemini 2.5 Flash, integrated IDEs, all-in-one platforms	Iterate on image + text overlay + composition adjustments in single interface without context switching
Cross-Domain Compositing	Blend elements from different sources, styles, or modalities into cohesive final assets	Layer-based tools, masked inpainting, style transfer, alpha compositing	Composite illustrated character into photoreal background via precise masked inpainting and lighting matching
Text Overlay Integration	Add typography, posters, labels, and multi-lingual text elements during generation phase	Ideogram, Qwen native text rendering, prompt-based typography control	Prompt: `text: "LAUNCH 2026" in bold serif, bottom-center, gold foil texture with subtle drop shadow`
Post-Generation Mastery Checklist	Systematically verify technical quality, aesthetic coherence, and prompt fidelity before delivery	Error detection workflow, quality gates, peer review protocols	Check hands/feet anatomy, verify eye direction consistency, confirm lighting coherence, test at target resolution, review negative prompt coverage, validate aspect ratio for platform, confirm color grading consistency, document seed/reference for reproducibility

Production Deployment Skills (Enterprise)

Sub-Skill	Explanation	Tools & Techniques
Model Serving	Deploy models via APIs with proper load balancing, latency management, and scaling	REST/gRPC endpoints, Kubernetes orchestration, load balancers, auto-scaling groups
Quantization & Optimization	Reduce model size and improve inference speed without significant quality loss	ONNX export, TensorRT, GGUF quantization, pruning, distillation techniques
Monitoring & Logging	Track errors, response times, token usage, and model drift in production environments	Prometheus/Grafana dashboards, custom telemetry, LangSmith, structured logging
CI/CD for AI Pipelines	Automate testing, validation, and redeployment of GenAI applications and workflows	GitHub Actions, MLflow, model versioning, automated regression testing, canary deployments

Connects to: All prior categories (refinement layer), Agent Orchestration (deployment automation and scaling)

6. AI Agent Orchestration (New for 2026)

Designing and managing multi-step, autonomous workflows where AI agents collaborate to achieve complex creative and operational goals.

Sub-Skill	Explanation	Tools & Techniques	Example
Agent Architecture Design	Understand how LLMs call tools, plan execution steps, and maintain contextual memory across interactions	ReAct pattern, Plan-and-Execute, memory buffers, tool-calling schemas	Agent workflow: `search web, summarize findings, generate image, draft social post, schedule publication`
Router Pattern Implementation	Direct incoming queries to appropriate specialized agents using semantic routing and intent classification	Embedding-based routing, intent classifiers, fallback mechanisms, load-aware dispatch	Query "create video ad for product launch" routed to VideoAgent; "write blog post" routed to TextAgent
SLM Optimization	Deploy 1B-10B parameter Small Language Models for cost-effective routine tasks with fallback to larger models	Model selection matrices, task routing logic, performance/cost benchmarking, hybrid inference	Use 7B SLM for email drafting and summarization; reserve 70B model for complex strategy and creative direction
Role-Based Agent Teams	Create specialized agents with distinct responsibilities that collaborate like a human creative team	PlannerAgent, ResearcherAgent, WriterAgent, ReviewerAgent, EditorAgent with clear handoff protocols	PlannerAgent breaks brief into tasks, ResearcherAgent gathers references, WriterAgent drafts, ReviewerAgent validates
Cross-Agent Communication	Design protocols for agent coordination, context sharing, and conflict resolution in multi-agent systems	Shared memory stores, message buses, structured JSON protocols, pub/sub patterns, state synchronization	Agents publish/subscribe to `task_context` channel; use consensus voting for creative decisions
Autonomous Research Workflows	Build agents that can independently search, evaluate, summarize, and organize findings from multiple sources	Web search tools, credibility scoring, summarization pipelines, knowledge graph construction	ResearchAgent tasked: "Gather Q1 2026 AI video generation trends from 5+ sources, output markdown report with citations"
Private Skill Deployment	Encode proprietary workflows, compliance requirements, and internal policies as private, auditable skills packages	SKILL.md format, policy engines, sandboxed execution, access controls, encryption	`compliance-check.skill` validates all outputs against company brand guidelines, legal requirements, and accessibility standards
Auditable Agent Workflows	Ensure every agent action, decision, and output is tracked, reviewable, and compliant with regulations	Immutable logs, decision tracing, human-in-the-loop checkpoints, regulatory reporting hooks	Log: `Agent X called tool Y at Z time with input A, output B, human approval C, final deployment D`
Sovereign AI Infrastructure	Deploy agent systems on private infrastructure with full data control and no external dependencies	On-premise deployments, VPC-isolated environments, air-gapped systems, private model hosting	Run agent cluster in private cloud; all data stays within organizational boundaries; zero external API calls for sensitive workflows
Skills Packaging (SKILL.md)	Write structured, portable, progressive-disclosure knowledge packages that encode workflows for AI agents	Modular sections, platform-agnostic syntax, validation rules, versioning, cross-platform compatibility	This document is a SKILL.md example; agents load only relevant sections per task via progressive disclosure

Connects to: All prior categories (orchestration layer); enables scaling of individual skills into production-ready, auditable, business-critical systems

7. LoRA Ecosystem & Parameter-Efficient Fine-Tuning (New for 2026)

Bridging prompt engineering and full model ownership through efficient adaptation techniques. LoRA enables targeted customization of foundation models for specific styles, subjects, and workflows without retraining the entire model.

Sub-Skill	Explanation	Tools & Techniques	Example
LoRA Fundamentals	Low-rank decomposition of weight updates: W = W₀ + BA where B and A are low-rank matrices injected into transformer layers	PEFT library, HuggingFace Diffusers, ComfyUI LoRA loader	`r=16, lora_alpha=32, target_modules=["to_q","to_k","to_v"]`
QLoRA Optimization	4-bit NF4 quantization of base model weights with LoRA adapters trained on top, enabling consumer GPU fine-tuning	bitsandbytes NF4, PEFT + transformers, Unsloth	Fine-tune 7B model on single 12GB GPU with 4-bit quantization + LoRA rank 16
DoRA (Weight-Decomposed)	Decompose weight updates into magnitude and direction components for higher fidelity style preservation	DoRA-PEFT, research implementations	`decompose=True` in LoRA config for detail-sensitive portrait style adaptation
AdaLoRA (Adaptive Rank)	Automatically allocate rank budgets per layer based on importance scoring, optimizing compute under constraints	AdaLoRA-PEFT, SVD-based importance	`rank_budget=0.5` — auto-allocates higher rank to critical attention layers
Rank-Stabilized LoRA (rsLoRA)	Rescale LoRA gradients by 1/√r to stabilize training at higher ranks, achieving closer-to-full-finetuning quality	rsLoRA-PEFT, custom scaling	`rscale=True` — enables rank 64+ without training instability
LoRA+ (Differential LR)	Apply different learning rates to A and B matrices in the low-rank decomposition for faster convergence	LoRA+ optimizer wrapper	`lr_A=1e-4, lr_B=1e-3` — B matrix learns 10× faster for better optimization
VeRA (Vector-based)	Share frozen random projection matrices across all layers with only per-layer scaling vectors trained	VeRA-PEFT, extreme compression	Train only 0.01% of parameters using shared frozen random bases
GLoRA (Global Prompt)	Global prompt-conditioned adapter with per-layer LoRA, enabling multi-task adaptation from a single model	GLoRA-PEFT, prompt tuning hybrid	`task_descriptor → global_adapter → per_layer_lora` for style switching
LoRA Style Libraries	Curate and combine multiple LoRA adapters (style, subject, lighting) at inference with weighted blending	CivitAI, HuggingFace Hub, ComfyUI multi-LoRA	`style_noir@0.6 + lighting_rembrandt@0.4 + subject_ceo@0.8`
LoRA Training Pipeline	End-to-end workflow: dataset curation → captioning → training → evaluation → deployment	OneTrainer, kohya-ss, sd-scripts, PEFT	20-50 curated images → QLoRA rank 16 → 10-20 epochs → safetensors export
S-LoRA Serving	Serve thousands of LoRA adapters on a single GPU with unified paging, tensor parallelism, and dynamic batching	S-LoRA, vLLM with LoRA, ComfyUI	Load 50+ LoRA adapters dynamically per request with KV cache sharing
LoRA × Framework Integration	Apply LoRA techniques across all framework domains: learned lighting patterns, character consistency adapters, negation-embedding LoRAs	Cross-domain adapter composition	Character LoRA × style LoRA × lighting LoRA combined for brand campaign consistency

Connects to: Prompt Engineering (learned patterns replace keywords), Photographic Literacy (LoRAs for specific lens/lighting), Identity Preservation (character LoRAs), Agent Orchestration (dynamic LoRA loading per task via S-LoRA)

Skill Synergy Matrix

Primary Skill	Reinforces	Enables	Critical For
Prompt Engineering	Photographic Literacy, Agent Orchestration	All downstream skills; precise creative control	Foundation of all AI interaction; campaign briefs
Photographic Literacy	Prompt Engineering, Skin Mastery	Realistic output; professional visual quality	Commercial photography; film pre-vis; brand assets
Strategic Negation	Identity Preservation, Post-Processing	Clean outputs; reduced revision cycles; artifact prevention	Character consistency; brand safety; client deliverables
Identity Preservation	Negation, Agent Orchestration	Multi-scene narratives; serialized content; IP development	Series content; brand campaigns; transmedia storytelling
Post-Processing	All prior skills	Production-ready deliverables; client-ready polish	Publication standards; commercial work; final assets
Agent Orchestration	All skills (scaling layer)	Autonomous workflows; enterprise systems; team augmentation	Production deployment; operational scaling; business integration
LoRA Ecosystem	Prompt Engineering, Identity Preservation	Learned visual patterns; character consistency; modular style libraries	Style customization; subject fine-tuning; multi-adapter serving

Pro Tip: Practice Skill Chaining

Don't learn skills in isolation. Build workflows that chain competencies:

Prompt Engineering -> Photographic Literacy -> Negation -> Consistency -> Post-Processing -> LoRA Optimization -> Orchestration

Example chain for a brand campaign:

Scaffold prompt with photographic vocabulary
Apply Rembrandt lighting + 85mm lens specs
Negate plastic skin + airbrushing
Lock character via --cref + --seed
Refine hands via inpainting; upscale with Topaz
Train character LoRA for consistent brand identity across unlimited assets
Deploy via agent workflow that dynamically loads style LoRAs for A/B variants

Competency Progression Path

Level	Key Capabilities	Typical Tools	Success Metric
Beginner	Natural language prompts; basic platform navigation; simple image generation	DALL-E 3, Gemini, ChatGPT, basic Midjourney	Generates coherent single images from simple prompts; understands basic platform UI
Intermediate	Technical vocabulary; lighting/lens control; basic negation; simple consistency techniques	Midjourney, SDXL, Leonardo, Ideogram, basic ComfyUI	Consistently produces on-brand, technically sound outputs; fixes common artifacts
Advanced	Seed/reference control; inpainting/outpainting; multi-tool hybrid workflows; video consistency	ComfyUI, Midjourney v6/v7, ControlNet, Topaz, Veo/Runway	Maintains character/style consistency across 10+ generations; ships multi-scene narratives
Expert	Agent system design; LoRA training & serving; production deployment; reusable skills packaging; auditable workflows	LangChain, PEFT, S-LoRA, vLLM, private model serving, SLMs, Kubernetes	Ships auditable, scalable AI systems; trains custom LoRAs; mentors others; contributes to frameworks

Quick-Start Templates (Copy-Paste Ready)

Image Generation Template

[Subject: specific, active] + [Action/pose] + [Lighting: pattern + direction] + [Lens: focal length + aperture] + [Style: artistic reference] + [Quality: native resolution + rendering terms] + [Negatives: weighted exclusions]

Example:
"Professional headshot of a software engineer, smiling naturally with genuine expression, Rembrandt lighting with key light 45deg high and subtle fill, 85mm lens f/2.8 shallow depth of field, photorealistic style referencing Annie Leibovitz, 4K native resolution with subsurface scattering on skin, (plastic skin:1.3) (airbrushed:1.2) (symmetrical face:1.1)"

Cinematic Video Template

[Scene description] + [Camera: movement + framing + lens] + [Lighting: time of day + pattern + motivation] + [Technical: fps + resolution + codec] + [Continuity: character/setting references] + [Negatives: drift prevention]

Example:
"Protagonist discovers crucial clue in dim library, camera: slow dolly zoom from wide establishing shot to tight close-up on face revealing realization, lighting: golden hour backlight through window with practical desk lamp fill, motivated shadows, 24fps 4K native ProRes, --cref character_sheet.jpg --sref film_grain_style --seed 8821, (facial drift:1.4) (background flicker:1.3) (inconsistent props:1.2)"

Agent Task Template

## Task: Research & Summarize AI Video Trends Q1 2026
**Agent Roles**: Planner -> Researcher -> Writer -> Reviewer -> Publisher
**Tools**: Web search API, summarization LLM (SLM), markdown formatter, brand compliance checker

**Validation Rules**:
- Sources: >=3 reputable tech publications (TechCrunch, Wired, arXiv)
- Length: 300-500 words executive summary + bullet points
- Tone: Professional, accessible, brand-aligned
- Compliance: No unverified claims; cite all statistics

**Output Format**: Markdown with H2 headers, bullet points, source links, key takeaways box

**Fallback Protocol**: If web search fails or source credibility <80%, use cached knowledge base + flag for human review + log incident

**Success Metric**: Report approved by ReviewerAgent with >=90% compliance score; published to internal wiki within 15 min of task initiation

Iterative Refinement Workflow

# 1. Generate base image with seed lock
prompt="corporate headshot, Rembrandt lighting, 85mm f/2.8" --seed 12345 --ar 4:5

# 2. Branch variant: change lighting only
prompt="corporate headshot, butterfly lighting, 85mm f/2.8" --seed 12345 --ar 4:5

# 3. Branch variant: change expression only
prompt="corporate headshot, smiling naturally, Rembrandt lighting, 85mm f/2.8" --seed 12345 --ar 4:5

# 4. Inpaint fix: repair hands if needed
--inpaint-mask hands --prompt "natural hand pose, relaxed fingers"

# 5. Upscale & grade
topaz-upscale --model photo-v4 --lightroom-preset "cinematic-portrait"

Platform & Tool Reference

Category	Recommended Tools	Best For	Learning Path
Foundation Image	Midjourney, SDXL, DALL-E 3	Rapid prototyping, concept art, learning fundamentals	Start here: master scaffold method + basic negation
Pro Image	Flux, Ideogram, Recraft, Qwen Image 2.0	Commercial work, text rendering, brand consistency, native 4K	Add: reference tools, multi-reference consistency, unified editing
Video Generation	Veo 3.1, Wan 2.6, Sora, Runway Gen-3	Cinematic sequences, social content, narrative storytelling	Layer: camera movement language, temporal consistency, VSA
Hybrid Editing	Photoshop (Gen Fill), ComfyUI, Topaz Photo AI, ACE++	Refinement, compositing, upscaling, professional finishing	Master: inpainting workflows, cross-domain compositing, external enhancement
Agent Frameworks	LangChain, AutoGen, CrewAI, custom builders, SKILL.md parsers	Workflow automation, multi-step tasks, team augmentation	Advance: router patterns, role-based teams, auditable workflows
Deployment	vLLM, TensorRT, ONNX Runtime, GGUF, Kubernetes	Production serving, optimization, cost management	Expert: quantization strategies, monitoring, CI/CD pipelines
Monitoring	Prometheus/Grafana, LangSmith, Arize, custom telemetry	Observability, drift detection, compliance auditing	Expert: implement full audit trails, automated alerting, regression testing

Deployment & Hosting (2026)

Best Free Frontend Deployment Stack

Platform	Free Tier	Best For
Cloudflare Pages	Unlimited requests/bandwidth	High traffic, edge performance
Vercel	Unlimited bandwidth	Next.js/React (best DX)
Netlify	100GB/month	JAMstack, forms, serverless
GitHub Pages	Unlimited (public repos)	Simple static sites
Surge.sh	Free CLI deploy	Quick prototyping

Quick Recommendations

Next.js/React -> Vercel (best integration)
Static/vanilla -> Cloudflare Pages (unlimited free)
Maximum free -> Cloudflare Pages + Turso (SQLite)
Quick CLI -> npx surge

Maintenance & Evolution Protocol

This framework is a living document. To keep skills current and compounding:

Quarterly Skill Audit
- Revisit emerging techniques (new model capabilities, research papers)
- Assess personal/team proficiency against competency matrix
- Update templates and examples with platform changes
Tool Rotation Practice
- Test 1-2 new platforms or techniques per quarter
- Retire outdated workflows; document migration paths
- Share findings with team via updated SKILL.md sections
Community & Research Sync
- Follow key sources: arXiv (cs.CV, cs.LG), platform changelogs, creator communities
- Participate in skill-sharing: contribute templates, report edge cases, validate techniques
- Attend workshops/webinars on emerging orchestration patterns
Document Contribution Loop
- When you discover a new technique or refinement, document it in SKILL.md format
- Use progressive disclosure: add to relevant section with clear connections to existing skills
- Version control: tag updates, maintain changelog, deprecate obsolete guidance
Synergy Review
- Every 6 months, map how new skills connect to existing categories
- Identify emerging synergy opportunities (e.g., new negation technique + agent workflow)
- Refactor framework structure if new category emerges (as Agent Orchestration did in 2026)
Release Integrity (learned in v5.0 → v5.1)
- A release plan is a hypothesis: re-derive remote state before acting (git ls-remote, gh pr list, gh api …/pages) — a stale transcript will otherwise describe a repo that no longer exists
- Verify ancestry on full history: in a depth-1 clone git merge-base and --is-ancestor answer from a truncated graph, so "already merged" and "nothing to push" both look plausible
- Derive, never restate: version, date, and counts in generated artifacts come from this document's own marker, and every release regenerates wiki.html + skills.pdf via tools/, then runs bash tools/session-restore.sh (gates G1–G13)
- Tag the version that is actually on master; a number in a plan is not a release

"Mastery isn't knowing every tool -- it's understanding how skills compound to solve real problems, and having the framework to learn the next one faster." -- AI Practitioner's Mantra, 2026

Appendix: Platform-Specific Syntax Quick Reference

Midjourney

# Character reference
--cref URL --cw 80

# Style reference
--sref URL

# Seed locking
--seed 12345

# Aspect ratio
--ar 16:9, --ar 9:16

# Negative prompting
--no plastic airbrushed cartoon

# Version control
--v 6.0, --style raw

Stable Diffusion / ComfyUI

# IP-Adapter reference
{"type": "ip-adapter", "image": "char.jpg", "weight": 0.8}

# ControlNet guidance
{"module": "openpose", "model": "control_v11p_sd15_openpose"}

# Negative embedding
negative_prompt = "(plastic skin:1.3), (extra fingers:1.4), cartoon"

# Seed and steps
{"seed": 12345, "steps": 30, "cfg": 7.0}

Agent SKILL.md Structure

## [Skill Name]
**Category**: [1-6]
**Version**: 1.0
**Last Tested**: [date]
**Platforms**: [Midjourney, SDXL, etc.]

### Purpose
[One-sentence goal]

### Inputs
- [Required parameters]
- [Optional parameters]

### Workflow
1. [Step one]
2. [Step two]

### Validation
- [Success criteria]
- [Failure handling]

### Connections
- Reinforces: [Skill X]
- Enables: [Skill Y]

License: CC-BY-SA 4.0 -- Share, adapt, and contribute improvements back to the community.

Source synthesis: AI Skills Research Addendum + Professional AI Creator Guide (Feb 2026).

Optimized for: Human readability + AI agent parsing (structured headings, consistent tables, explicit connectors, progressive disclosure).

Format: Save as SKILL.md in project root, GitHub wiki, or AI agent knowledge base.

aiskills-photog

About