Replicate black-forest-labs/flux-2-pro

Black Forest Labs sells FLUX.2 in tiers, and Pro is the one aimed at people with real deadlines - the KB's Flux 2 panel describes the family as the strongest open-weights image generation on paper, with the Dev and Klein lines carrying the open-weights flag while Pro and Max sit at the commercial top. What Pro buys you on Replicate, via this node, is that top tier's generation on their GPUs instead of the 18–24GB of VRAM the full stack wants locally. On a CPU-only ComfyUI box, "professional tier FLUX" isn't otherwise a sentence you can say.

Honestly, the node is near-identical to the Max node in this pack - same schema shape, same knobs - because the two are siblings in the same commercial line. prompt in, IMAGE out (tensor), API_JSON alongside. Inputs that matter:

• aspect_ratio (default "1:1", eleven choices) - including match_input_image and custom. Pick custom to take over width and height directly (both 256–2048, default 0 meaning the model decides).
• resolution (default "1 MP", or match_input_image / 0.5 / 2 / 4 MP) - the quality-tier dial. Pro's whole point is fidelity, so 2–4 MP is where it earns its keep - and where the bill climbs.
• safety_tolerance (default 2, range 1–5) - Replicate's filter strength. 1 strictest, 5 most permissive. If a run gets refused and you didn't know this existed, now you do.
• seed (default 0) - reproducibility, since Replicate is stateless.
• output_format (webp default, or jpg/png) and output_quality (80).
• IMAGE_1 through IMAGE_8 - up to eight reference images. This is the family's generational change: a 24B VLM text encoder that reasons over references in a way Flux 1's CLIP+T5 couldn't. Character consistency, style transfer, multi-image composition - the Pro workflows live here.
• dry_run / force_rerun - test free, and force a fresh call past ComfyUI's cache.

The pack-wide wisdom applies with extra force here: this is the expensive one. Pro at 4 MP with eight references is not an iteration tier. The sane pattern is to scout on Klein (the pack's Replicate black-forest-labs/flux-2-klein-9b node) at 1 MP, then graduate to Pro for the finals - same family's quality direction, far cheaper to explore. Dry-run your payload, eyeball the API_JSON, and check the invoice on a test batch before you run a real production run.

Install is the shared pack flow - ComfyUI Manager (search ComfyUI-API-DockerCPU), or:

cd ComfyUI/custom_nodes
git clone https://github.com/trustypangolin/ComfyUI-API-DockerCPU
cd ComfyUI-API-DockerCPU
pip install -r requirements.txt

Set REPLICATE_API_TOKEN and restart. It's under 🎨 DockerCPU API/🎨 Replicate.

One more thing worth saying: don't get hung up on the tier labels. The KB's Flux 2 panel is clear that benchmark-winning and community-winning are different games, and for a lot of jobs the cheaper Klein node will be indistinguishable in output. Pro is for when you specifically need the top tier's fidelity and can justify the price per image. If you can't articulate why you need Pro, you probably need Klein.