NS WaveSpeed Flux Kontext Max
Flux Kontext Max, Minus the Knobs
- client
- output_image
Some nodes give you forty knobs. This one gives you four, and that's the point. NSWaveSpeedFluxKontextMax is the top-of-the-line Flux Kontext variant on WaveSpeed's hosted API, wrapped with a deliberately minimal interface: prompt, input image, guidance, and a safety setting. It's the "I know what I want, just render it" node of the family - and its default prompt, "To toy style", telegraphs the use case: transforming an image into a stylized render (toy, clay, cartoon) with the subject intact.
Compare it to the pack's Kontext Dev node, which exposes steps, seed, output format, and num_images. Max strips all that down to the essentials and hands the rest to the model. If you spend your life tweaking steps and seeds, this will feel like it's hiding things from you. If you want a one-shot style transform with the least friction, this is the one to grab.
How it works
It calls WaveSpeed's flux-kontext-max endpoint with your image_url and prompt. The inputs are:
- image_url - a URL string (not an IMAGE tensor) from an upload/URL-producing node. Same gotcha as the rest of the WaveSpeed family.
- prompt - the transformation instruction. Default
"To toy style", but the model is happy with "make it a watercolor painting," "convert to a cyberpunk render," and so on. - guidance_scale - 1–20, default 3.5. How strictly it follows the prompt. This is your one quality lever.
- safety_tolerance - 1 (strict) to 5 (permissive), default 2. It's the provider-side content filter; raise it only if you're getting false-positive refusals on content you're allowed to generate.
- enable_sync_mode - wait for the render to finish before returning, on by default.
Output: output_image as an IMAGE tensor. There's no seed, no step count, no format choice - the endpoint decides, and sync mode hides the async dance.
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 in Settings → Symbiotica, or WAVESPEED_API_KEY) into client.
Troubleshooting
- The prompt gets ignored - Kontext Max is style-focused; vague prompts ("make it better") don't steer it. Be explicit about the look. Also remember guidance at 3.5 is moderate - nudge it up for stricter adherence.
- Refusals on innocent content -
safety_tolerancedefault is 2. If you're getting false positives, raising it to 3–4 is the intended fix. There's no abliterating a hosted model - this is the only filter lever you get. - The result is a style applied to a scene you didn't want - the input image dictates the subject. Check
image_urlactually points at the image you think it does (a stale URL is a classic). - No steps/seed - can I make results reproducible? - not from this node; that's the deal you made for the minimal surface. If you need seed control, the Dev node is the one with the knobs.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| client | WAVESPEED_AI_API_CLIENT | — | |
| prompt | STRING | To toy style | Text prompt describing the desired transformation or scene |
| image_url | STRING | The image URL to transform (connect from Upload Image node) | |
| guidance_scale | FLOAT | 3.51–20 | How closely to follow the prompt (1.0 = loose, 20.0 = strict) |
| safety_tolerance | COMBO | 2 | Safety filter tolerance level (1 = strict, 5 = permissive) |
| enable_sync_mode | BOOLEAN | true | Wait for image generation to complete before returning |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| output_image | IMAGE | — |