🤖 Leon FLUX Image API

Leon FLUX Image API is the Swiss-army FLUX node of the pack: one node, six FLUX 1.1-era models behind a single model_choice dropdown, all routed through the HyprLab API. From FLUX 1.1 Pro Ultra down to FLUX Schnell, plus the FLUX Pro Canny structure-guided variant and FLUX Krea Dev for image-to-image - you swap models by changing one dropdown, and every model reuses the same socket layout. It's the node you reach for when you want FLUX quality without a local checkpoint and without rebuilding your graph for each variant.

The catch, and it's the same one as the rest of the pack's API nodes: this is a hosted, paid service. Nothing runs locally. You pay per generation, the node posts your prompt to api.hyprlab.io/v1/images/generations (the api_url default, which you can repoint), and converts the returned image into an RGBA tensor for your graph.

How it works - the model-aware payload

Here's the part that makes this node more than a dumb wrapper: it only sends the parameters that each model actually accepts. Read the tooltips, because they're precise about applicability.

• FLUX 1.1 Pro Ultra - the flagship. Uses image_prompt_strength (0 to 1), aspect_ratio, and raw. Feed it an image for guidance and it blends it in at the strength you set.
• FLUX 1.1 Pro / FLUX Pro Canny / FLUX Dev / FLUX Schnell - these use steps, height, and width. Height and width must be multiples of 32 (the node validates this and will raise a clear error otherwise; min 256, max 1440). Canny is the structure one - it wants a Canny edge map as the image prompt.
• FLUX Krea Dev - the image-to-image specialist. Uses prompt_strength, num_inference_steps, and guidance.

Image guidance comes two ways: input_image_prompt_socket (a tensor) or input_image_prompt_url (a hosted URL). The base class will refuse to send both at once - "provide either an image tensor or an image URL, not both" - and it prefers the URL when given. A 256x256 minimum applies.

Outputs are the pack standard: image (tensor), image_url, seed.

Installing it

The pack-wide install:

cd ComfyUI/custom_nodes
git clone https://github.com/l3ony2k/comfyui-leon-nodes
pip install -r requirements.txt

Restart ComfyUI (or install via ComfyUI Manager, search "Leon"). Light dependencies, no model downloads - the models live on HyprLab's side.

Where people get burned

Three things trip people up. First, the key: YOUR_HYPRLAB_API_KEY is a placeholder, and a bad key fails slowly because of the 5-attempt backoff retry - check the console before you blame the node. Second, model-specific parameters are silently ignored on models that don't use them: setting steps on Ultra does nothing, setting raw on Dev does nothing. The tooltips are your map; trust them. Third, the 32-multiple height/width rule - the node hard-errors on it, which is friendlier than most API wrappers but still catches people used to freeform dimensions. If you need FLUX 2's multi-image remixing instead of single-image guidance, the FLUX 2 node in this same pack is the one to use.