[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-80091":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":15,"stars7d":17,"stars30d":18,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":32,"readmeContent":33,"aiSummary":34,"trendingCount":16,"starSnapshotCount":16,"syncStatus":17,"lastSyncTime":35,"discoverSource":36},80091,"HiDream_O1-ComfyUI","Saganaki22\u002FHiDream_O1-ComfyUI","Saganaki22","HiDream O1 Image nodes + LoRA training for ComfyUI ","https:\u002F\u002Fgithub.com\u002FHiDream-ai\u002FHiDream-O1-Image",null,"Python",63,11,58,1,0,2,5,3,46.74,"MIT License",false,"main",[25,26,27,28,29,30,31],"comfyui","comfyui-nodes","image-editing","image-to-image","lora","lora-training","text-to-image","2026-06-12 04:01:26","# HiDream_O1-ComfyUI\n\n**HiDream O1 Image nodes for ComfyUI** — local HiDream O1 generation with text prompts, optional reference images, BF16\u002FFP16\u002FFP32\u002FFP8 model loading, FlashAttention, SageAttention, preview updates, and ComfyUI DynamicVRAM\u002FAimdo integration.\n\n[![Demo](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDemo-HiDream--O1--Image-green)](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FHiDream-ai\u002FHiDream-O1-Image)\n[![GitHub](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Saganaki22%2FHiDream__O1--ComfyUI-black)](https:\u002F\u002Fgithub.com\u002FSaganaki22\u002FHiDream_O1-ComfyUI)\n\n[中文文档](README_ZH.md)\n\n\u003Cimg width=\"2560\" height=\"1440\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fadfcbb51-6e04-4daf-82cf-99b2052f32de\" \u002F>\n\n\n## Features\n\n- HiDream O1 Image generation directly inside ComfyUI\n- Text-only and reference-image workflows\n- Dynamic `image_1` to `image_12` inputs on the sampler node\n- Optional Dev layout conditioning via JSON bbox input\n- `keep_image1_aspect` toggle for reference-driven output aspect ratio\n- BF16, FP16, FP32, FP8 E4M3FN, and FP8 E5M2 loader options\n- FP8 mixed-weight loading using ComfyUI manual-cast style compute\n- FlashAttention, SageAttention, and PyTorch SDPA attention backends\n- Progress previews through ComfyUI's sampler progress bar\n- Dev\u002FDev-2604 patch-grid smoothing node for reducing visible tile seams\n- AI Toolkit-aligned HiDream O1 LoRA training nodes\n- ComfyUI model management, unload, DynamicVRAM, and Aimdo\u002FVBAR support\n\n\n\u003Cimg width=\"2010\" height=\"899\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F116f408b-dcce-4e01-b8e9-93566f8a2cca\" \u002F>\n\n\n## Installation\n\n### Method 1: ComfyUI Manager\n\nSearch for `HiDream O1` or `HiDream_O1-ComfyUI` in ComfyUI Manager and install it.\n\n### Method 2: Manual Install\n\n```bash\ncd ComfyUI\u002Fcustom_nodes\ngit clone https:\u002F\u002Fgithub.com\u002FSaganaki22\u002FHiDream_O1-ComfyUI.git\ncd HiDream_O1-ComfyUI\npython -m pip install -r requirements.txt\n```\n\nRestart ComfyUI after installing or updating.\n\n**Suggested `transformers` version: 4.57.1 – 5.3** (newer versions may break compatibility).\n\nHiDream's May 13, 2026 upstream update notes that PyTorch 2.9.x is not recommended because of a Qwen3-VL issue. This node logs a warning when it detects 2.9.x.\n\n## Model Setup\n\nDownload the complete model folder from one of the links below and place it inside `ComfyUI\u002Fmodels\u002Fdiffusion_models\u002F`:\n\n| Precision | VRAM | Download |\n|-----------|------|----------|\n| Full BF16 | ~18–20 GB | [drbaph\u002FHiDream-O1-Image-BF16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-BF16) |\n| Full FP16 | ~18–20 GB | [drbaph\u002FHiDream-O1-Image-FP16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-FP16) |\n| Full FP8 | ~10–11 GB | [drbaph\u002FHiDream-O1-Image-FP8](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-FP8) |\n| Dev 2604 BF16 | ~18–20 GB | [drbaph\u002FHiDream-O1-Image-Dev-2604-BF16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-BF16) |\n| Dev 2604 FP16 | ~18–20 GB | [drbaph\u002FHiDream-O1-Image-Dev-2604-FP16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-FP16) |\n| Dev 2604 FP8 | ~10–11 GB | [drbaph\u002FHiDream-O1-Image-Dev-2604-FP8](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-FP8) |\n| Dev BF16 | ~18–20 GB | [drbaph\u002FHiDream-O1-Image-Dev-BF16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-BF16) |\n| Dev FP16 | ~18–20 GB | [drbaph\u002FHiDream-O1-Image-Dev-FP16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-FP16) |\n| Dev FP8 | ~10–11 GB | [drbaph\u002FHiDream-O1-Image-Dev-FP8](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-FP8) |\n\n**Example — FP8 (lowest VRAM):**\n\n1. Go to [drbaph\u002FHiDream-O1-Image-FP8](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-FP8)\n2. Download the **entire model folder** (all files, not just the safetensors)\n3. Place it at `ComfyUI\u002Fmodels\u002Fdiffusion_models\u002FHiDream-O1-Image-fp8\u002F`\n\nThe folder must contain the full Hugging Face support files:\n\n```text\nconfig.json\nchat_template.json\ngeneration_config.json\npreprocessor_config.json\ntokenizer.json\ntokenizer_config.json\nvocab.json\nmerges.txt\nmodel.safetensors\n```\n\nThe original sharded format also works if the folder contains `model.safetensors.index.json` and all shard files.\n\nThe model loader always shows the built-in converted model choices: Full\u002FDev BF16, FP16, FP8, plus Dev-2604 BF16, FP16, and FP8. If the selected model already exists locally, it is used. If it is missing, enable `download_if_missing` and the selected model will be downloaded into `ComfyUI\u002Fmodels\u002Fdiffusion_models`.\n\nLocal folder matching is case-insensitive, so `HiDream-O1-Image-Dev-FP8`, `hidream-o1-image-dev-fp8`, and the default target folder casing all resolve to the same built-in choice. The loader dropdown only shows the built-in HiDream O1 model choices.\n\n### Upstream Artifact Note\n\nThe original\u002Ffull HiDream O1 model can show grid artifacts or other reference-image artifacts. In the upstream issue tracker, a HiDream developer recommends trying the Dev model because it should have fewer grid artifacts, and notes that reference-image generation is still being improved: [HiDream-ai\u002FHiDream-O1-Image issue #1](https:\u002F\u002Fgithub.com\u002FHiDream-ai\u002FHiDream-O1-Image\u002Fissues\u002F1#issuecomment-4412738522).\n\nIn general, the Full model is the better choice for realism and photographic detail. The Dev model is faster and often better for illustration, digital design, and cleaner grid\u002Fartifact behavior, but it can be more sensitive to scheduler and resolution choices.\n\n| Variant | Precision | Hugging Face repo | Target folder |\n|---------|-----------|-------------------|---------------|\n| Full | `auto`, `bf16`, `fp32` | [`drbaph\u002FHiDream-O1-Image-BF16`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-BF16) | `HiDream-O1-Image-bf16` |\n| Full | `fp16` | [`drbaph\u002FHiDream-O1-Image-FP16`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-FP16) | `HiDream-O1-Image-fp16` |\n| Full | `fp8_e4m3fn`, `fp8_e5m2` | [`drbaph\u002FHiDream-O1-Image-FP8`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-FP8) | `HiDream-O1-Image-fp8` |\n| Dev 2604 | `auto`, `bf16`, `fp32` | [`drbaph\u002FHiDream-O1-Image-Dev-2604-BF16`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-BF16) | `HiDream-O1-Image-Dev-2604-bf16` |\n| Dev 2604 | `fp16` | [`drbaph\u002FHiDream-O1-Image-Dev-2604-FP16`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-FP16) | `HiDream-O1-Image-Dev-2604-fp16` |\n| Dev 2604 | `fp8_e4m3fn`, `fp8_e5m2` | [`drbaph\u002FHiDream-O1-Image-Dev-2604-FP8`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-FP8) | `HiDream-O1-Image-Dev-2604-fp8` |\n| Dev | `auto`, `bf16`, `fp32` | [`drbaph\u002FHiDream-O1-Image-Dev-BF16`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-BF16) | `HiDream-O1-Image-Dev-bf16` |\n| Dev | `fp16` | [`drbaph\u002FHiDream-O1-Image-Dev-FP16`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-FP16) | `HiDream-O1-Image-Dev-fp16` |\n| Dev | `fp8_e4m3fn`, `fp8_e5m2` | [`drbaph\u002FHiDream-O1-Image-Dev-FP8`](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-FP8) | `HiDream-O1-Image-Dev-fp8` |\n\n## Nodes\n\n### HiDream O1 Model Loader\n\nLoads a local HiDream O1 model folder and returns a Comfy-managed model handle.\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `model_name` | `HiDream-O1-Image-BF16` | Built-in HiDream O1 model choice |\n| `precision` | `auto` | Detects safetensors dtype, or forces `bf16`, `fp16`, `fp32`, `fp8_e4m3fn`, `fp8_e5m2` |\n| `attention` | `auto` | `auto`, `flash`, `sdpa`, or `sage` |\n| `download_if_missing` | `false` | Downloads the selected built-in model if it is not installed locally |\n\n### HiDream O1 Conditioning\n\nCreates prompt conditioning for the sampler.\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `prompt` | cinematic portrait prompt | Text instruction for generation |\n| `enhanced_prompt` | optional input | Optional `STRING` input from ComfyUI's bundled `Prompt Enhance` subgraph or any prompt-enhancer output; when connected and non-empty, it replaces the prompt textbox |\n| `negative_prompt` | empty | Negative prompt used as the unconditional CFG branch in full mode when `guidance_scale` is above `1.0`; dev mode ignores CFG |\n\nOptional bundled ComfyUI prompt-enhancement flow (generic, not HiDream-O1-specific):\n\n```text\nPrompt Enhance -> HiDream O1 Conditioning enhanced_prompt\n```\n\nComfyUI's bundled `Prompt Enhance` blueprint is a generic subgraph around the `Google Gemini` node, not part of the native HiDream-O1 model\u002Fconditioning path and not the local Gemma 4 `Generate Text` node. The generic `Generate Text` node can still be used if you provide your own instruction prompt, but it is not the same prompt-enhancement workflow.\n\n### HiDream O1 LoRA\n\nApplies a LoRA between the model loader and sampler:\n\n```text\nHiDream O1 Model Loader -> HiDream O1 LoRA -> HiDream O1 Sampler\n```\n\nThe LoRA dropdown reads from `ComfyUI\u002Fmodels\u002Floras\u002F`, including supported LoRA files inside symlinked folders.\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `lora_name` | `None` when no LoRAs are found | LoRA file |\n| `strength` | `1.0` | Model strength from `-10.0` to `10.0`; `0` disables the LoRA |\n\n### HiDream O1 Dev Smoothing\n\nApplies patch-grid smoothing between the model loader or LoRA node and the sampler:\n\n```text\nHiDream O1 Model Loader -> HiDream O1 Dev Smoothing -> HiDream O1 Sampler\nHiDream O1 Model Loader -> HiDream O1 LoRA -> HiDream O1 Dev Smoothing -> HiDream O1 Sampler\n```\n\nThis node is gated to Dev and Dev-2604 model folders. It runs extra shifted patch predictions during the last denoise steps and blends them back into the latent patch grid to reduce visible seams.\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `steps` | `4` | Final denoise steps to smooth; `0` disables smoothing |\n| `strength` | `0.5` | Blend strength for shifted patch prediction |\n| `schedule` | `constant` | Strength schedule over smoothing steps |\n| `shift_mode` | `rotate` | Patch-grid shift pattern |\n| `adaptive_threshold` | `0.0` | Skip smoothing when estimated seam intensity is below this value; `0` disables skipping |\n| `multiscale` | `false` | Adds a smaller patch-grid offset |\n| `cfg_aware` | `false` | Also smooths the unconditional branch when CFG is active; costs extra forwards |\n\n### HiDream O1 LoRA Training\n\nExperimental text-to-image LoRA training is available directly inside ComfyUI:\n\n```text\nHiDream O1 Dataset Maker -> HiDream O1 Train Config -> HiDream O1 LoRA Trainer\n```\n\nThe trainer is for image\u002Fcaption datasets only. Reference-image, edit, and subject-personalization training are not wired yet.\n\nDataset folder layout:\n\n```text\nmy_dataset\u002F\n  image_001.png\n  image_001.txt\n  image_002.jpg\n  image_002.txt\n```\n\nEach `.txt` file should contain the caption for the image with the same basename. The Dataset Maker writes a `train.jsonl` manifest that the trainer consumes.\n\nTraining notes:\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `base_model_name` | `HiDream-O1-Image-BF16` | Full O1 BF16 weights |\n| `resolution` | `1024` | Images are resized\u002Fcropped to a patch-aligned training size |\n| `target_preset` | `aitoolkit` | Trains linear-like layers except `lm_head`, `patch_embed`, and `visual`, matching AI Toolkit's O1 ignore list |\n| `loss_target` | `velocity` | Converts the model's x0 prediction into flow velocity before loss |\n| `noise_scale` | `8.0` | Scales training noise the same way as AI Toolkit's HiDream O1 flow scheduler |\n| `timestep_type` | `linear` | AI Toolkit's O1 default |\n| `max_loss` | `1.0` | Caps extreme loss spikes like AI Toolkit's O1 default |\n| `lora_rank` \u002F `lora_alpha` | `32` \u002F `32` | AI Toolkit-style linear LoRA defaults |\n| `weight_decay` | `0.0001` | AdamW weight decay default from AI Toolkit's job config |\n| `save_dtype` | `bf16` | LoRA checkpoint tensor dtype |\n| `max_steps` | `3000` | Total training steps |\n| `save_every_steps` | `250` | Checkpoint interval |\n\nOutputs are saved under `ComfyUI\u002Fmodels\u002Floras\u002F\u003Coutput_name>\u002F` as `.safetensors` files plus `hidream_o1_lora_config.json`. After training, select the saved `.safetensors` in the normal **HiDream O1 LoRA** node.\n\nThe trainer follows AI Toolkit's May 2026 HiDream O1 recipe: it adds scaled noise with `noise_scale=8.0`, feeds the noisy image patches through the Qwen-VL model, converts the x0 prediction into a velocity-equivalent prediction, and trains against `noise * noise_scale - image`. The trainer runs in-process and blocks the ComfyUI queue while it is active. Use the Full model for training; Dev is intentionally not exposed in the trainer because it is distilled and may train unpredictably.\n\nFor a deeper setup and tuning guide, see [HiDream O1 training notes](docs\u002FHIDREAM_O1_TRAINING.md).\n\n### HiDream O1 Sampler\n\nRuns the model and outputs a ComfyUI `IMAGE`.\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `model_type` | `auto` | Uses `dev` settings if the model folder name contains `dev`, otherwise full settings |\n| `width` | `2048` | Requested output width; internally snapped to a supported patch-aligned resolution |\n| `height` | `2048` | Requested output height; internally snapped to a supported patch-aligned resolution |\n| `steps` | `0` | `0` means auto: 50 for full; dev always uses the upstream fixed 28-step schedule |\n| `seed` | `42` | Random seed |\n| `guidance_scale` | `5.0` | CFG scale for full mode; dev mode ignores CFG |\n| `shift` | `-1.0` | `-1` means auto: 3.0 for full, 1.0 for dev |\n| `noise_scale_start` | `7.5` | Initial noise scale |\n| `noise_scale_end` | `7.5` | Final noise scale |\n| `noise_clip_std` | `2.5` | Noise clipping standard deviation |\n| `dev_editing_scheduler` | `flow_match` | Dev edit mode scheduler when exactly one reference image is connected; `flash` remains available |\n| `layout_bboxes` | empty | Optional JSON string or JSON file path for layout conditioning with reference images |\n| `preview_every` | `4` | Sends a decoded preview every N steps; `0` disables previews |\n| `keep_image1_aspect` | `false` | Only applies when `image_1` is connected |\n| `force_offload` | `false` | Unloads the model immediately after generation |\n| `image` | `0` | Dynamic reference image count, from `0` to `12` |\n\nReference image inputs are optional. Set `image` to `0` for text-only generation, or increase it to show `image_1`, `image_2`, and so on up to `image_12`.\n\n## Precision Notes\n\n`auto` detects the model storage dtype from the safetensors file. For native mixed FP8 folders, the large matrix weights should be `float8_e4m3fn` while small tensors such as norms and biases stay BF16\u002FFP16.\n\nDo not set `config.json` to `float8_e4m3fn`. Transformers may try to use FP8 as PyTorch's global default dtype, which fails. Keep config dtype as `bfloat16`; this node detects FP8 from the safetensors tensors themselves.\n\nThe loader exposes the normal FP8 options only.\n\n## Scheduler\n\nThe sampler automatically picks the scheduler based on model type:\n\n| Model type | Scheduler | Notes |\n|------------|-----------|-------|\n| Full (`auto`) | `FlowUniPCMultistepScheduler` | Higher-order solver, generates more detail |\n| Dev text \u002F subject | `FlashFlowMatchEulerDiscreteScheduler` | Custom Euler with built-in noise scaling, tuned for fewer steps |\n| Dev edit with one reference | `FlowMatchEulerDiscreteScheduler` by default | Matches the May 13, 2026 upstream Dev editing scheduler update; `flash` is still selectable |\n\nWhen `model_type` is `auto`, the folder name is checked for `dev` — if not found, the full model path is used with UniPC.\n\nDev follows the upstream recipe: fixed 28-step timetable, guidance `0.0`, shift `1.0`, and noise defaults `7.5 \u002F 7.5 \u002F 2.5` when using `flash`. If dev images look noisy, oddly colored, or washed out near the last few steps, reset `noise_scale_start`, `noise_scale_end`, and `noise_clip_std` to those defaults, use the `flash` or `auto` attention backend, and pin the output to one of the internal supported resolutions: `2048x2048`, `2304x1728`, `1728x2304`, `2560x1440`, `1440x2560`, `2496x1664`, `1664x2496`, `3104x1312`, `1312x3104`, `2304x1792`, or `1792x2304`. Upstream recommends the Full model for editing tasks.\n\n## Attention Backends\n\n| Option | Description |\n|--------|-------------|\n| `auto` | Uses FlashAttention when available, otherwise SDPA |\n| `flash` | Requires FlashAttention [Optimal] |\n| `sage` | Requires the `sageattention` package [Not Optimal] |\n| `sdpa` | Uses PyTorch scaled dot-product attention |\n\n## Links\n\n- Demo: [HiDream-O1-Image](https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FHiDream-ai\u002FHiDream-O1-Image)\n- Dev 2604 BF16 model: [drbaph\u002FHiDream-O1-Image-Dev-2604-BF16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-BF16)\n- Dev 2604 FP16 model: [drbaph\u002FHiDream-O1-Image-Dev-2604-FP16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-FP16)\n- Dev 2604 FP8 model: [drbaph\u002FHiDream-O1-Image-Dev-2604-FP8](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-2604-FP8)\n- BF16 model: [drbaph\u002FHiDream-O1-Image-BF16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-BF16)\n- FP16 model: [drbaph\u002FHiDream-O1-Image-FP16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-FP16)\n- FP8 model: [drbaph\u002FHiDream-O1-Image-FP8](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-FP8)\n- Dev BF16 model: [drbaph\u002FHiDream-O1-Image-Dev-BF16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-BF16)\n- Dev FP16 model: [drbaph\u002FHiDream-O1-Image-Dev-FP16](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-FP16)\n- Dev FP8 model: [drbaph\u002FHiDream-O1-Image-Dev-FP8](https:\u002F\u002Fhuggingface.co\u002Fdrbaph\u002FHiDream-O1-Image-Dev-FP8)\n- Upstream project: [HiDream-ai\u002FHiDream-O1-Image](https:\u002F\u002Fgithub.com\u002FHiDream-ai\u002FHiDream-O1-Image)\n- Node repository: [Saganaki22\u002FHiDream_O1-ComfyUI](https:\u002F\u002Fgithub.com\u002FSaganaki22\u002FHiDream_O1-ComfyUI)\n\n## License\n\nThis custom node is released under the MIT License. The HiDream O1 model has its own license and usage terms; check the upstream Hugging Face model page before redistribution or commercial use.\n","HiDream_O1-ComfyUI 是一个为 ComfyUI 提供图像生成和 LoRA 训练功能的插件。其核心功能包括基于文本提示和参考图像的本地 HiDream O1 图像生成，支持多种精度模型（BF16\u002FFP16\u002FFP32\u002FFP8）加载，以及 FlashAttention 和 SageAttention 等先进的注意力机制。此外，它还集成了 ComfyUI 的 DynamicVRAM 和 Aimdo 功能以优化内存使用，并提供进度预览。该插件适用于需要高质量图像生成与编辑、尤其是结合了自然语言处理能力的应用场景，如创意设计、数字艺术创作等。","2026-06-11 03:59:12","CREATED_QUERY"]