🤖 Leon Stable Diffusion 3 Ultra API

sd3-ultra is Stability's top-end SD3 tier - the version of the MMDiT family that's billed for quality over speed, with the SD3-era strengths (good prompt adherence, solid text rendering) pushed as far as the architecture goes. It's also a model most people will never run locally: the SD3 line never got a healthy open ecosystem going, the launch was rough enough that the weights got pulled from CivitAI, and Ultra specifically is a paid API product. Leon_StableDiffusion_3_Ultra_API_Node is the pack's wrapper for exactly that: one model, hardcoded in the payload (sd3-ultra, no dropdown - there's only one choice), one HyprLab key, and the generation happens on Stability's side while your GPU does nothing.

Honest framing, because it matters: if you want SD3-family results and you're cost-averse, the sibling Leon_StableDiffusion_35_API_Node gives you a Medium option that's much cheaper per call. This node exists for the job where you specifically want the Ultra tier.

How it works

Same StableDiffusionImageGenerationNodeBase as the pack's other SD nodes: build an OpenAI-style payload, POST to https://api.hyprlab.io/v1/images/generations with a Bearer token, retry with exponential backoff (up to five attempts), and decode the response into an RGBA tensor. The payload carries model: "sd3-ultra", your prompt, negative_prompt, aspect_ratio, output_format, and response_format. The prompt field supports (word:weight) weighting, per the tooltip - the same syntax as the SD 3.5 node.

Inputs that matter

• prompt - the workhorse. With only one model in the dropdown and no cfg/sampler knobs, the prompt (and its (word:weight) hints) is nearly your whole control surface.
• negative_prompt - what not to include. Underrated on this node because there's nothing else to tune.
• aspect_ratio - nine options (16:9, 1:1, 21:9, 2:3, 3:2, 4:5, 5:4, 9:16, 9:21).

Outputs are the pack standard: image, image_url, seed - with the seed capped at 4,294,967,294 like the other SD3 nodes.

Installing

ComfyUI Manager (search Leon) or clone:

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

Restart, paste a HyprLab key. Nothing downloads; the only hardware involved is someone else's.

Common issues

The usual API-node failures apply - swap the default YOUR_API_KEY_HERE, and read the console when you get a 4xx, since the node prints the full error body. The Ultra-specific gotcha is expectations: because there's no sampler or CFG field, prompt quality does all the work, so if results feel soft, tighten the prompt and lean on (word:weight) before you blame the node. And remember it's metered and it's the expensive tier - if you're iterating on an idea, validate the prompt on a cheaper node first, then spend the Ultra credits on the keeper.