ai-asset-generator
Skill1 starsChanged 3 months agoMITView on GitHub
AI 2D asset generation for websites, social, and client deliverables — product photography, hero images, lifestyle shots, textures, mockups. Routes photoreal vs illustration vs edit-in-place to the right model; worked-example vendor is the Higgsfield CLI (FLUX, Flux Kontext, Nano Banana) with fal.ai fallback and Midjourney hand-off — swap your own. Triggers on "generate an image", "product photo", "hero image", "mockup", "color variants", or when a website builder needs assets. NOT for video — use social-video-producer.
- Reads credentials
In this repository47 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- access-transfer-trackerSkill
- ai-3d-generatorSkill
- ai-demo-builderSkill
- apple-grade-designSkill
- brand-guide-builderSkill
- client-onboardingSkill
- cold-email-campaignSkill
- cold-emailSkill
- company-teardownSkill
- competitive-ads-extractorSkill
- content-calendarSkill
- content-repurposerSkill
- content-research-writerSkill
- content-strategySkill
- daily-briefSkill
- discovery-call-prepSkill
- email-composerSkill
- follow-up-nurtureSkill
- gbp-local-seoSkill
- hormozi-100m-leadsSkill
- interactive-website-builderSkill
- invoice-trackerSkill
- lead-flow-verifierSkill
- lead-scraperSkill
- mcp-integrationSkill
- meeting-listenerSkill
- post-purchase-email-seriesSkill
- productization-journalSkill
- proposal-generatorSkill
- prospect-workupSkill
- regulated-copy-complianceSkill
- retainer-reportingSkill
- review-managerSkill
- scroll-demo-builderSkill
- seo-auditSkill
- site-launch-cutoverSkill
- site-qa-checklistSkill
- social-media-auditorSkill
- social-media-managerSkill
- social-video-producerSkill
- stylized-3d-worldbuildingSkill
- video-analyzerSkill
- web-researchSkill
- website-builderSkill
- weekly-client-checkinSkill
- welcome-email-seriesSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 77
Website hero assets: read clients/<slug>/brand/brand.md first → hero via Midjourney hand-off or your quality model → section textures …
- Very specific: line 102
Image-to-video prompts describe motion only: When an asset from this skill feeds video generation: prompts under 30 words, …
- Very specific: line 117
Check clients/<slug>/brand/brand.md before any brand-adjacent imagery.
- Distinctive: line 118
Never generate images of real, named people.
- Distinctive: line 122
Keep the best working prompt per client in their brand folder for reuse.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AI Asset Generator
Generate production-quality 2D assets for websites, social, proposals, and deliverables. Read this entire file before generating. The vendor stack below is the worked example — swap in your own tools; the routing logic and doctrine are vendor-independent.
Model Routing — the decision logic
Route by JOB, not by habit. The categories that matter:
What are you generating?
├── Product photography (e-commerce, clean background)?
│ ├── Have an existing product photo? → EDIT-IN-PLACE model (e.g. Flux Kontext):
│ │ background swaps, color variants — keep everything else identical
│ └── No photo? → PHOTOREAL text-to-image model (e.g. FLUX-class)
├── Hero / campaign imagery (artistic, mood-driven)?
│ └── Cheapest quality-tier stylized model; Midjourney hand-off if the operator
│ has a subscription (hero picks benefit from a human eye on variations)
├── Image that must contain TEXT (signage, packaging, logo mockup)?
│ └── The strongest text-rendering model available (e.g. Nano Banana class)
│ — ALWAYS visually verify the rendered text before shipping
├── Recurring character across a campaign?
│ └── Train a reusable character reference once; reuse it per generation
├── Quick mockup / start+end frames for video work? → fast cheap model
├── Copyright-indemnified imagery (nervous client)? → commercial-safe vendor
│ (e.g. Adobe Firefly) — propose the subscription first if not connected
├── Batch (50+ variants)? → price the batch FIRST, then a scripted loop
├── Background removal / upscale? → dedicated utility models, not prompting
└── Texture / seamless tile? → photoreal model + tiling prompt (below)Before every job: query the model's current params, then price the job before generating (higgsfield model get <job_type> / higgsfield generate cost ... in the worked example). Per-image prices drift — never quote from memory. Anything in your vendor CLI that DEPLOYS to vendor infrastructure (site/app generators) is a deploy: explicit operator green-light required.
Prompting Patterns
Universal structure — completeness matters more than order:
[Subject] + [Composition/framing] + [Lighting] + [Style/mood] + [Technical specs] + [Negative constraints]Clean product shot:
[Product with specific material, color, finish, shape].
Single product, centered, white seamless studio background.
Soft diffused lighting from above-left, subtle contact shadow.
Professional e-commerce product photography, sharp focus, high resolution.
No props, no text, no watermarks, no humans.Lifestyle shot: product placed naturally in a specific setting, one camera angle, natural time-of-day lighting, shallow depth of field, "no text overlays, no logos, photorealistic".
Color variants (highest-value e-commerce play): with an edit-in-place model —
Change the color of the [product] from [current] to [target].
Keep everything else identical: same lighting, background, angle, shadows.
Photorealistic, no other changes.Photograph once, generate the other 6-8 colorways — saves the client thousands in studio time. Calculate that ROI when pricing the service (the operator sets the number).
Hero images: visually striking with deliberate negative space for the headline — specify WHERE ("clear negative space on the left third for headline text"), dramatic lighting, "no text, no watermarks, no faces", aspect ratio matched to placement (16:9 hero, 1:1 social, 9:16 stories).
Textures: "Seamless tileable [material]. [Palette]. Subtle variation, not perfectly uniform. Flat lighting, no perspective distortion. High resolution, seamless edges." — "seamless tileable" + "flat lighting" are the load-bearing modifiers.
Midjourney hand-off (no API): your job is the perfect prompt in a copy-friendly code block with all parameters appended (--ar matched to use, --s 250 balanced / 500+ artistic / 100 literal, --raw for photoreal, low --c for client work). The operator generates, picks, upscales, saves; you optimize for web afterward. When options are wanted, vary mood via --s, not by rewriting the subject.
Workflow Patterns
- Product photography package: client sends 2-5 phone photos → background removal → studio shots (photoreal model) → color variants (edit-in-place) → 3-5 lifestyle placements → deliver at required resolution (typically 2000x2000).
- Website hero assets: read
clients/<slug>/brand/brand.mdfirst → hero via Midjourney hand-off or your quality model → section textures automated → optimize everything for web. - Social batch: define the calendar → consistent branded backgrounds → per-post images in one style/palette → text overlays via Pillow/HTML by default (bake text in only with the text-strong model, then verify visually).
- Proposal visuals: match the prospect's real industry and brand colors woven into the prompt — never generic imagery.
Web Optimization (every image, every time)
Raw AI images run 2-10MB — never ship them. Convert to WebP (sharp -i in.png -o out.webp --quality 85), generate responsive sizes (480/768/1200/1920), serve via <picture> + srcset with loading="lazy", explicit width/height.
| Use case | Max width | Target size |
| Hero image | 1920px | <200KB |
| Section background | 1920px | <150KB |
| E-commerce product | 2000px | <300KB |
| Thumbnail | 480px | <50KB |
| Social post | 1080px | <200KB |
File organization: clients/<slug>/site/assets/img/{hero,products,products/variants,lifestyle,textures,social}/. Own ventures: ventures/<slug>/brand/img/.
Gotchas Carried From Production
- Flatten transparency before generating from a PNG. Transparent inputs behave unpredictably in image-to-image and image-to-video models — composite onto the target background color first. Same rule when a still will later feed an image-to-video pipeline.
- Image-to-video prompts describe motion only. When an asset from this skill feeds video generation: prompts under 30 words, motion/lighting/atmosphere only — never re-describe the image contents.
- Text in images is guilty until proven innocent. Any generated image containing rendered text gets downloaded and visually Read before shipping — models misspell.
- Never ship any generated or stock image without viewing it. Download it, Read it, confirm subject, framing, and that it actually looks good.
Anti-Slop Rules
- Over-saturated color → add "natural color grading" / "desaturated"
- Plastic skin → avoid people unless needed; "natural skin texture, imperfections, no airbrushing"
- Perfect symmetry → real scenes have subtle asymmetry
- Stock-photo poses → be specific about context and action
- "HDR" in prompts → use "natural dynamic range" or "film-like contrast"
- Purple/blue AI gradients → the universal tell; use the client's actual brand palette
Hard Rules
- Check
clients/<slug>/brand/brand.mdbefore any brand-adjacent imagery. - Never generate images of real, named people; never AI-generate client headshots/team photos without explicit client approval; disclose AI generation if the client asks.
- Optimize every image for web before delivery — no 4MB PNGs on websites.
- API keys live in
.envonly — reference by variable name, never paste values. - Price every batch before running it; log significant generation spend in your decision log.
- Keep the best working prompt per client in their brand folder for reuse.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

