flux3-klein-stills
Installation
SKILL.md
flux3-klein-stills
Run every graph with the bundled client: python scripts/comfy_client.py --url $COMFY_URL run workflows/<file> --set NODE.INPUT=VALUE --out out/. Upload input files first with python scripts/comfy_client.py upload <file>. Before the first run on a server, check it with python scripts/preflight.py --url $COMFY_URL workflows/<file>. Node packs, model files, VRAM and weight licenses for this skill are in references/requirements.md. Use open-weights models only, and never swap in a hosted or closed model.
Pick a graph
| Need | Workflow | Set |
|---|---|---|
| Default. Distilled FLUX 3 Klein, 16 steps at CFG 1 | flux3_klein_distilled_t2i.api.json |
4.text=prompt, 7.seed, 6.width/height |
| FLUX 3 Klein base with a Qwen prompt upsampler (50 steps, CFG 3.5; slower, and a negative prompt does work at this CFG) | flux3_klein_t2i.api.json |
18:17.value=prompt, 18:7.seed. The bbox/layout nodes only feed a PreviewAny, so they do not steer the sampler |
| The same base graph as a clean template | flux3_klein_t2i_template.api.json |
18:17.value, 18:7.seed |
| FLUX.2 Klein 9B | flux2_klein_9b_t2i.api.json |
6.text, 14.noise_seed, and width/height on nodes 5 and 13 (keep them equal) |
| Lean drafts (FLUX.2 Klein 4B) | flux2_klein_t2i_lean.api.json |
74.text, 73.noise_seed, and width/height on nodes 62 and 66 |
For detailed prompt rules, see video-director/references/quality-playbook.md.