ai-image-editing
The AI image-editing router — inpainting/object removal, background removal, upscaling, outpainting, old-photo restoration, and retouch, routed task-first to…
The FLUX craft skill (Black Forest Labs) — generate and edit on-brand images with the right variant and license. Use when someone wants to generate images with FLUX/FLUX.2/Kontext, edit a generated image, keep a character or product consistent across a campaign, render legible
$ npx -y skills add social-media-skills/skills --skill flux --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/fluxContext preview
The summary Claude sees to decide when to auto-load this skill.
The FLUX craft skill (Black Forest Labs) — generate and edit on-brand images with the right variant and license. Use when someone wants to generate images with FLUX/FLUX.2/Kontext, edit a generated image, keep a character or product consistent across a campaign, render legible
name: flux description: >- The FLUX craft skill (Black Forest Labs) — generate and edit on-brand images with the right variant and license. Use when someone wants to generate images with FLUX/FLUX.2/Kontext, edit a generated image, keep a character or product consistent across a campaign, render legible text in images, get brand-exact colors, pick between FLUX variants, or asks if their FLUX use is commercially licensed. Uses the PIXEL framework. Reads image-prompt + brand-profile + design-and-templates first. The agent writes prompts/edit instructions and can call the API where connected; the HUMAN judges every image; WoopSocial publishes. License spine: [dev] outputs are commercial-OK but self-hosting for a commercial service needs a paid BFL tier; Apache-2.0 paths are [schnell]/[klein] 4B. Never use unpermitted likeness, clone trade dress, strip provenance, or invent stats. Distinct from image-prompt, ideogram/nano-banana/Midjourney (sibling tools), ai-image-editing (the edit router this feeds), and canva. version: 1.0.0
The **FLUX image tool skill** — pick the variant + license, instruct in scenes, lock references + edit instead of re-rolling, evaluate deliberately, and label before publishing. The agent prompts (and can call the API where connected); the **human judges every image**; **WoopSocial publishes**. (Ships with `tools/integrations/flux.md`.)
FLUX's 2026 edge isn't just quality; it's **control**: multi-reference consistency (up to ~8–10 images in one call — a campaign-consistent character or product with **no fine-tuning**), in-context editing ("change the jacket, keep everything else"), **hex-code brand colors** as parameters, 32k-token scene prompts, and typography clean enough that ad headlines are production-viable. Two top-1% edges most users miss. **(1) The license split is the trap:** [dev] **outputs** are commercially usable, but **self-hosting [dev] to serve commercial work** (clients, a product) needs a paid BFL tier — the agency tier includes just **3 clients** before per-client fees — and the dev license **requires content filters or manual review**, which BFL says it may verify at random. Apache-2.0 freedom lives in [schnell] and [klein] 4B; the hosted APIs are the easiest commercial path (license + signed provenance handled). **(2) Edit, don't re-roll:** a 95%-right image is one Kontext-style instruction from done — re-rolling throws away the 95%. And the craft shift: FLUX reads **natural-language scene briefs**, not tag soup — describe subject, light, mood, camera; put exact in-image text in quotes and **verify every character.**
1. **image-prompt** — the model-agnostic prompt craft + router above all image tools. 2. **brand-profile** + **design-and-templates** — the system (colors → hex codes, type, logo rules).
(Depth: `references/the-pixel-framework.md`.)
self-host; [dev] = outputs OK / services need a paid tier + required filters; [max] for web-grounded; verify at bfl.ai.
brand-exact color; exact text in quotes; prompt upsampling helps [dev].
in-context edits fix the 5% instead of re-rolling; chain small edits, watch drift; verify rendered text.
composition lands; side-by-side drift review on sets.
strip the signed provenance metadata; then WoopSocial publishes.
FLUX.2 (Nov 2025, current flagship): 32B rectified-flow + Mistral-3 VLM, multi-reference (~8–10 images), 4MP, clean small-size typography (the "text soup" era largely fixed), hex-color parameters, photorealism reducing "the AI look"; [klein] (Jan 2026) = sub-second on consumer GPUs (~13GB VRAM for the 4B), **4B Apache 2.0** (the 9B klein is non-commercial); [max] adds real-time **web-grounded generation**; FLUX.1 Kontext = the in-context editing line (a selectable partner model in Photoshop Beta's Generative Fill). Licensing (bfl.ai, attributed): dev outputs commercial-OK; self-hosted commercial services need paid tiers (developer ~10K img/mo single-domain, not client work; agency ~100K/mo, 3 clients included); **filters or manual review required on [dev]** with random verification stated; API applies **cryptographically-signed provenance metadata**; non-removable CSAM/NCII filters on API. Runs via Playground → APIs (BFL/fal/Replicate/ Together/Cloudflare) → self-host (ComfyUI; FLUX.2 [dev] quantized on an RTX 4090). **Attribute all; verify-quarterly.** Full detail: `references/flux-2026-reality.md`. The variant table, scene-prompt pattern, consistency + edit workflows, and two worked examples: `references/prompt-patterns-and-templates.md`.
otherwise); the **human judges every image** (no fabricated "that looks great"; side-by-side set reviews); **WoopSocial publishes** finished exports — it does **not** generate or edit media.
filter/review obligation; Apache-2.0 = [schnell]/[klein] 4B; high-stakes → counsel (not legal advice). **Likeness/IP/provenance:** no unpermitted real-person likeness (real or AI lookalike), no cloned trade dress or copyrighted characters, never strip provenance, **AI-disclosure** where required
Give your AI agent the skills of a top-1% social media team. 106 of them, free.
The AI image-editing router — inpainting/object removal, background removal, upscaling, outpainting, old-photo restoration, and retouch, routed task-first to…
The AI music + sound-design skill for social -- original/licensed audio beds and sound design for Reels/TikToks/Shorts/videos. Use when someone needs…
Use to get a brand and its content CITED and RECOMMENDED by AI answer engines — the GEO (Generative Engine Optimization) / AI-search-visibility skill. Run when…
The model-agnostic AI-video router and brief — the counterpart to image-prompt. Use when someone asks "which AI video tool should I use," "make an AI video,"…
The AI narration / voiceover mini-skill (ElevenLabs-led). Use when someone wants an "AI voiceover," "narration," "text-to-speech for a video," "voice for my…
Social media analytics and reporting — read native platform data honestly and turn it into next actions. Use when someone wants to "check my analytics," "see…