Card: link the ComfyUI nodes and single-file weights

README.md

@@ -41,6 +41,7 @@ Built by [LogoLabs](https://logolabs.org), which makes AI logo generation, as a
 | **GenEval** (official) | **0.550**. SDXL 0.55, SD 2.1 0.50, PixArt-α 0.48, SD 1.5 0.43 (published) |
 | **Qwen-Image-Bench** (1,000 prompts) | **28.2**, against SD 1.5's 29.1; on the Pareto frontier among open models of similar size |
 | **Speed** | 1.9 s per image on an RTX 4060 (50 steps); also runs in the browser: [WebGPU demo](https://huggingface.co/spaces/Logolabs/agate-webgpu) |
+| **ComfyUI** | [logolabs/agate-comfyui](https://github.com/logolabs/agate-comfyui): 1.7 s per image, standard SD 1.5 latents, single-file weights in [`comfyui/`](https://huggingface.co/Logolabs/agate-preview-001/tree/main/comfyui) |
 | **Strengths** | Placing objects (left of, on top of), binding colours to objects, faces, styles, flat logos, long prompts |
 | **Weaknesses** | Exact text, counts above three, negation, anything above 256 px; not yet converged |
 | **Licence** | MIT, for code and weights |
@@ -125,6 +126,22 @@ image.save("teapot.png")
 On an RTX 4060, one image takes 1.9 s (2.9 s with `autoguide`). At 191M parameters, kernel-launch overhead
 costs more time than the arithmetic, and CUDA graphs remove most of it.
 
+### ComfyUI
+
+Nodes: **[logolabs/agate-comfyui](https://github.com/logolabs/agate-comfyui)**. In ComfyUI Manager use *Install via Git URL*,
+or `git clone` the repo into `ComfyUI/custom_nodes` and `pip install -r requirements.txt`.
+
+- **Agate Loader** reads a single-file checkpoint from `ComfyUI/models/agate/`. The official
+  [`agate-preview-001.safetensors`](https://huggingface.co/Logolabs/agate-preview-001/blob/main/comfyui/agate-preview-001.safetensors)
+  (522 MB, bf16, tokenizer and config inside) downloads automatically if it is missing.
+- **Agate Sampler** outputs a standard SD 1.5 latent, so the stock *VAE Decode* works, and it chains into upscalers
+  and img2img: it takes a latent and a `denoise` strength. It shows live previews and works with ComfyUI's
+  memory management, including `--lowvram`.
+- **Agate Generate** is the all-in-one node that returns an image directly.
+- **Speed**, measured on an RTX 4060 with previews on: 1.7 s per image at 50 steps, 1.05 s at 30 steps (a good
+  draft setting), 4.6 s for a batch of four.
+- **Example workflows** for txt2img, all-in-one generation and a 4× upscale ship in the repo.
+
 ## Why Agate
 
 **Why we built it.** [LogoLabs](https://logolabs.org) builds AI logo generation. Agate is our attempt at a better