106 social media skills for AI agents - strategy, writing, video, design, platform growth, publishing, and analytics. Works with Claude, Cursor, OpenClaw, Hermes & 40+ agents.
FLUX (Black Forest Labs) — integration & connection guide
What it is: the image generation + in-context editing model family this stack uses for photorealistic,
typographically-clean, reference-consistent imagery. FLUX generates/edits; the human judges; WoopSocial
publishes. Skill: skills/flux/; the model-agnostic router is skills/image-prompt/. Variants and license
terms shift per release — verify-quarterly at bfl.ai, docs.bfl.ai, and the Hugging Face model cards.
Connection layers
- Hosted APIs (the easiest commercial path): BFL API (docs.bfl.ai) or third-party endpoints (fal, Replicate, Together, Cloudflare) — pay-per-image, commercial license handled, cryptographically-signed provenance metadata applied, non-removable CSAM/NCII filters (Hive/Microsoft). The agent can call these where an API key/connection exists; REST, async webhooks on some providers.
- BFL Playground: browser, zero-setup — for prompt iteration and evaluating fit before committing.
- Self-host (ComfyUI / diffusers): FLUX.2 [dev] 32B (quantized runs on an RTX 4090 w/ remote text encoder), [klein] 4B (~13GB VRAM, sub-second), FLUX.1 [schnell]/[dev]/Kontext [dev]. The agent supplies prompts/edit instructions; the human runs the pipeline.
The variant × license map (≈, verify-quarterly)
| Variant | Access | License |
|---|---|---|
| FLUX.2 [pro] / [flex] / [max] | API | Commercial via API terms ([max] adds web-grounded generation) |
| FLUX.2 [dev] (32B) | Open weights | Non-commercial for self-hosted services; OUTPUTS commercial-OK |
| FLUX.2 [klein] 4B | Open weights | Apache 2.0 (9B klein = non-commercial) |
| FLUX.1 [schnell] | Open weights | Apache 2.0 |
| FLUX.1 [dev] / Kontext [dev] | Open weights | Non-commercial (same outputs-vs-service line) |
Self-host commercial tiers (bfl.ai/licensing, attributed): Builder ≈ 10K img/mo, single-domain internal/marketing, not client work · Platform ≈ 100K/mo · Professional ≈ 100K/mo — up to 3 client domains included, per-client fees beyond · Enterprise = custom. The [dev] license requires content filters or manual review in deployment — BFL states it may approach known deployers at random to verify. High-stakes → counsel (not legal advice).
Capability notes (FLUX.2 era)
Multi-reference: up to ~8–10 images/call → campaign-consistent characters/products with no fine-tuning · in-context editing (Kontext line): targeted changes, rest intact, chain small edits · 32k-token prompts (scene briefs, not tags; [dev] benefits from prompt upsampling) · hex color codes as parameters (brand-exact) · clean typography (quote exact text; character-verify before shipping) · up to 4MP · photorealistic lighting/physics.
The stack workflow (canonical)
image-prompt (router/craft) + brand-profile/design-and-templates (hex codes, type) → FLUX generate →
multi-reference for sets → in-context edits for the last 5% → human judges (side-by-side drift review) → export →
scheduling-and-queue → WoopSocial (upload media → attach → validate → create post; per-platform required fields validate atomically — see woopsocial.md).
Hard lines
Human judges every render · rights verified for the variant actually used (outputs-vs-service line; filter obligation on [dev]) · no unpermitted real-person likeness (real or AI lookalike) · no cloned competitor ads/ trade dress or copyrighted characters · never strip provenance metadata · AI-disclosure where required (EU AI Act; C2PA) · rendered text character-verified · no invented statistics (data viz → infographic-and-data-viz rules) · never state a license/price/benchmark as fact without verification · WoopSocial does not generate or edit images.