NS WaveSpeed Flux Kontext Dev

"Turn pictures into anime style." "Make it a claymation." "Renders this photo as a Pixar still." That's the whole pitch of Flux Kontext Dev - an image-to-image model from Black Forest Labs that takes an input image and re-renders it in a style you describe, keeping the subject and composition intact. NSWaveSpeedFluxKontextDev is the ComfyUI wrapper: give it an image URL and a transformation prompt, get the styled image back.

It's hosted on WaveSpeed, so no local Flux install, no VRAM, no weights. And because it's image-to-image rather than text-to-image, the value is immediate and demo-friendly - point it at a product shot and ask for "studio advertising, soft gradient background" and you've got a campaign image in one call.

How it works

The node posts your image_url and prompt to WaveSpeed's flux-kontext-dev endpoint. The core input is the prompt - note the default is literally "Turn pictures into anime style", which tells you what this model is for. Style-transformation instructions read best as "make it X style" or "convert to Y look"; the model keeps the semantics and swaps the rendering.

The knobs:

• image_url - again a URL string, not an IMAGE tensor. Connect it from a node that outputs a URL (the tooltip's "Upload Image node").
• guidance_scale - 1–20, default 2.5. Dev runs low; 2.5 is already above the strictest end of text-to-image Flux. Raise it if the transformation is too loose about following your instruction.
• num_inference_steps - 10–100, default 28. More steps, more refinement, slower bill.
• seed (-1 = random), output_format (jpeg/png/webp), enable_sync_mode, and optional num_images (1–4) for a few candidates at once.

Output: output_image as an IMAGE tensor, ready to feed a save node or an upscaler.

Install and setup

Part of the Symbiotica pack. ComfyUI Manager → search "Symbiotica", or:

cd ComfyUI/custom_nodes
git clone https://github.com/symbiotica-ai/comfyui-nodes.git symbiotica
pip install -r symbiotica/requirements.txt

Restart, and wire a NSWaveSpeedClient (key via Settings → Symbiotica or WAVESPEED_API_KEY) into the client socket.

Troubleshooting

• IMAGE won't connect to image_url - it's a URL string input. Your upload node must output a URL; a raw IMAGE tensor will not plug in.
• The style barely changes - Dev at guidance 2.5 is conservative. Raise guidance_scale toward 5–8 and add more style vocabulary to the prompt ("anime, cel-shaded, Studio Ghibli palette") instead of a bare "anime."
• The subject gets mangled - too much guidance or too many steps. Style transfer should preserve the subject; if it doesn't, lower guidance and steps and lean harder on the prompt wording.
• Random results across runs - seed is -1 (random) by default. Set a seed to iterate reproducibly.