{ "display_name": "Ideogram 4.0", "model_name": "Ideogram 4.0", "version": "Base", "architecture": "ideogram", "family": "ideogram4", "author": "Ideogram", "description": "Ideogram 4.0 open-weight text-to-image model (9.3B single-stream DiT) run locally in ComfyUI. Best-in-class in-image text rendering and typography. Uses a Qwen3-VL-8B text encoder and the Flux 2 VAE, plus a second 'unconditional' transformer that drives dual-branch CFG (there is no user negative prompt). Authored in natural language. Weights are non-commercial; generated images may be used commercially.", "license": "Ideogram Non-Commercial Model Agreement (inference code Apache-2.0)", "tags": [ "ideogram", "ideogram4", "text-rendering", "typography", "open-weight", "consumer" ], "url": "https://huggingface.co/Comfy-Org/Ideogram-4", "release_date": "2026-06-03", "tag_sources": [], "tag_format": "spaces", "prompt_style": "natural", "trigger": "", "default_prompt_id": "ideogram-default", "negative": "", "notes": "REQUIRES a ComfyUI version with native Ideogram 4 support (the UNETLoader 'ideogram4' arch, CLIPLoader 'ideogram4' type, Ideogram4Scheduler, DualModelGuider, CFGOverride). Weights from the UN-GATED Comfy-Org/Ideogram-4 mirror (no HF token needed). Two transformers load via two UNETLoaders; the unconditional one feeds DualModelGuider.model_negative. Recommended: Default = 20 steps, Turbo = 12, Quality = 48; DualModelGuider guidance 7; CFGOverride [3, 0.7, 1]; scheduler mu 0.5 / std 1.75; sampler euler. Native output up to 2048px. On a 32 GB card (RTX 5090) full fp8 fits with text-encoder offload; drop the unconditional transformer or encoder to nvfp4 if VRAM is tight. NON-COMMERCIAL weights: fine for personal/local use; commercial deployment of the model needs Ideogram's paid license (generated images themselves are fine to use commercially, except training competing models).", "files": [ { "label": "Diffusion Model", "folder": "diffusion_models", "variants": [ { "precision": "fp8", "filename": "ideogram4_fp8_scaled.safetensors", "quick_hash": "e9380be7fe0165c7", "size_bytes": 9280741285, "source": { "provider": "huggingface", "repo": "Comfy-Org/Ideogram-4", "path": "diffusion_models/ideogram4_fp8_scaled.safetensors" }, "sha256": null }, { "precision": "nvfp4", "filename": "ideogram4_nvfp4_mixed.safetensors", "size_bytes": 5490000000, "source": { "provider": "huggingface", "repo": "Comfy-Org/Ideogram-4", "path": "diffusion_models/ideogram4_nvfp4_mixed.safetensors" }, "sha256": null } ] }, { "label": "Unconditional Transformer", "folder": "diffusion_models", "variants": [ { "precision": "fp8", "filename": "ideogram4_unconditional_fp8_scaled.safetensors", "quick_hash": "76383548b85c3b56", "size_bytes": 9280741293, "source": { "provider": "huggingface", "repo": "Comfy-Org/Ideogram-4", "path": "diffusion_models/ideogram4_unconditional_fp8_scaled.safetensors" }, "sha256": null }, { "precision": "nvfp4", "filename": "ideogram4_unconditional_nvfp4_mixed.safetensors", "size_bytes": 5490000000, "source": { "provider": "huggingface", "repo": "Comfy-Org/Ideogram-4", "path": "diffusion_models/ideogram4_unconditional_nvfp4_mixed.safetensors" }, "sha256": null } ] }, { "label": "Text Encoder (Qwen3-VL-8B)", "folder": "text_encoders", "variants": [ { "precision": "fp8", "filename": "qwen3vl_8b_fp8_scaled.safetensors", "size_bytes": 10588637512, "source": { "provider": "huggingface", "repo": "Comfy-Org/Ideogram-4", "path": "text_encoders/qwen3vl_8b_fp8_scaled.safetensors" }, "sha256": null }, { "precision": "nvfp4", "filename": "qwen3vl_8b_nvfp4.safetensors", "size_bytes": 6310000000, "source": { "provider": "huggingface", "repo": "Comfy-Org/Ideogram-4", "path": "text_encoders/qwen3vl_8b_nvfp4.safetensors" }, "sha256": null } ] }, { "label": "VAE", "folder": "vae", "filename": "flux2-vae.safetensors", "size_bytes": 336213556, "source": { "provider": "huggingface", "repo": "Comfy-Org/Ideogram-4", "path": "vae/flux2-vae.safetensors" }, "sha256": null } ], "nodes": { "Ideogram4Scheduler": { "steps": 20, "steps_range": [ 12, 48 ], "width": 1024, "height": 1024 }, "DualModelGuider": { "cfg": 7 }, "CFGOverride": { "cfg": 3, "start_percent": 0.7, "end_percent": 1.0 }, "EmptyFlux2LatentImage": { "width": 1024, "height": 1024, "_resolutions": [ { "width": 1024, "height": 1024 }, { "width": 896, "height": 1152 }, { "width": 1152, "height": 896 }, { "width": 832, "height": 1216 }, { "width": 1216, "height": 832 } ] } } }