> ## Documentation Index > Fetch the complete documentation index at: https://comfytv.org/llms.txt > Use this file to discover all available pages before exploring further. # 自带工作流所需的模型文件 ComfyTV 在 `workflows//` 下面预置了一套精选的 ComfyUI 工作流。每个工作流都引用了若干模型文件,**这些文件 ComfyUI 自己必须能在`ComfyUI/models/` 树下找到**。本页按工作流列出每个所需的文件和它们应该放在哪个子目录。 > 如果你想用**自己的** ComfyUI 工作流(换模型、换 sampler、换 LoRA),上面这些都不需要下 , 见 [custom-workflows.zh.md](/zh/guides/custom-workflows),里面说明了怎么把自定义工作流 JSON + `_preset.json` 绑定接进来。 ComfyUI 通常会在首次启动时自动建好所有子目录;缺哪个手动建一下即可。下面所有目录都是相对于 `ComfyUI/models/` 的相对路径。 *** ## Generate · 文本(Text) | 工作流 | 文件 | 目录 | | ------------------ | ----------------------- | ---------------- | | **Local Qwen3 4B** | `qwen_3_4b.safetensors` | `text_encoders/` | 下载:[https://huggingface.co/Comfy-Org/flux2-klein/tree/main/split\_files/text\_encoders](https://huggingface.co/Comfy-Org/flux2-klein/tree/main/split_files/text_encoders) *** ## Generate · 图像(Image) | 工作流 | 文件 | 目录 | | ------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------- | | **Local SD1.5** / **Local SD1.5 I2I** | `v1-5-pruned-emaonly-fp16.safetensors` | `checkpoints/` | | **Image Ideogram4 T2I** | `ideogram4_fp8_scaled.safetensors`、`ideogram4_unconditional_fp8_scaled.safetensors`、`qwen3vl_8b_fp8_scaled.safetensors`、`flux2-vae.safetensors` | `diffusion_models/`、`diffusion_models/`、`text_encoders/`、`vae/` | *** ## Generate · 视频(LTX 2.3) 4 个变体共用同一套 base 模型 + 文本编码器 + LoRA。 | 工作流 | 文件 | 目录 | | ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------- | | **Local LTX 2.3 T2V** | `ltx-2.3-22b-dev-fp8.safetensors`、`ltx-2.3-spatial-upscaler-x2-1.1.safetensors`、`ltx_2.3_22b_distilled_1.1_lora_dynamic_fro09_avg_rank_111_bf16.safetensors`、`gemma-3-12b-it-abliterated_lora_rank64_bf16.safetensors`、`gemma_3_12B_it_fp4_mixed.safetensors` | `checkpoints/`、`latent_upscale_models/`、`loras/`、`loras/`、`text_encoders/` | | **Local LTX 2.3 I2V** | 同 T2V | 同 | | **Local LTX 2.3 FLF2V** | `ltx-2.3-22b-distilled-fp8.safetensors`、`gemma_3_12B_it_fp4_mixed.safetensors` | `checkpoints/`、`text_encoders/` | | **Local LTX 2.3 IA2V** | 同 T2V/I2V | 同 | 下载: * [https://huggingface.co/Lightricks/LTX-2.3-fp8](https://huggingface.co/Lightricks/LTX-2.3-fp8) * [https://huggingface.co/Lightricks/LTX-2.3](https://huggingface.co/Lightricks/LTX-2.3) * [https://huggingface.co/Comfy-Org/ltx-2.3](https://huggingface.co/Comfy-Org/ltx-2.3) * [https://huggingface.co/Comfy-Org/ltx-2](https://huggingface.co/Comfy-Org/ltx-2) *** ## Generate · 音频(Audio) | 工作流 | 文件 | 目录 | | -------------------- | ------------------------------ | -------------- | | **ACE-Step v1 Song** | `ace_step_v1_3.5b.safetensors` | `checkpoints/` | ACE-Step v1 是 ComfyUI 核心原生支持,不需要装额外的 custom node。 *** ## 图像编辑(Edits) | 工作流 | 文件 | 目录 | | ----------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------ | | **Flux Canny Edit**(image-edit) | `flux1-canny-dev.safetensors`、`t5xxl_fp16.safetensors` *(或 `t5xxl_fp8_e4m3fn_scaled.safetensors`)*、`clip_l.safetensors`、`ae.safetensors` | `diffusion_models/`、`text_encoders/`、`text_encoders/`、`vae/` | | **Flux Fill Inpaint**(inpaint) | `flux1-fill-dev.safetensors`、t5 + clip\_l + ae(同上) | `diffusion_models/`、`text_encoders/`、`text_encoders/`、`vae/` | | **Fooocus SDXL Inpaint**(inpaint) | `juggernautXL_version6Rundiffusion.safetensors`、`fooocus_inpaint_head.pth` | `checkpoints/`、`inpaint/` | | **Flux Fill Outpaint**(outpaint) | 同 Flux Fill Inpaint | 同 | | **Fooocus SDXL Outpaint**(outpaint) | 同 Fooocus SDXL Inpaint | 同 | | **LaMa Erase**(erase) | `big-lama.pt` | `inpaint/` | | **BiRefNet Cutout**(cutout) | `birefnet.safetensors` | `background_removal/` | | **Ultrasharp 4x**(upscale) | `4x-UltraSharp.pth` | `upscale_models/` | 下载: * Flux dev:[https://huggingface.co/Comfy-Org/flux1-dev](https://huggingface.co/Comfy-Org/flux1-dev) * t5xxl / clip\_l:[https://huggingface.co/comfyanonymous/flux\_text\_encoders](https://huggingface.co/comfyanonymous/flux_text_encoders) * VAE(ae.safetensors):[https://huggingface.co/Comfy-Org/Lumina\_Image\_2.0\_Repackaged](https://huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged) * LaMa Erase:需要安装 [`Acly/comfyui-inpaint-nodes`](https://github.com/Acly/comfyui-inpaint-nodes) 来提供 `INPAINT_LoadInpaintModel` 节点 + 仓库里的 `big-lama.pt`。 * Fooocus inpaint head:Fooocus inpaint patch 自带(`fooocus_inpaint_head.pth`)。 * BiRefNet:ComfyUI 核心原生支持(`LoadBackgroundRemovalModel` + `RemoveBackground`);[https://huggingface.co/Comfy-Org/BiRefNet](https://huggingface.co/Comfy-Org/BiRefNet) *** ## 图像变体(单工作流 N 输出) "多视角"(Face / Product / Character / Multi-cam 9)和"序列" (Story 4 / Storyboard 25)两类 preset 共用同一套 Qwen base,**只是 LoRA 不同**。 | 组 | 文件 | 目录 | | --------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------- | | **Multi-cam / 3-view 套** | `qwen_image_edit_2511_fp8mixed.safetensors`、`Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors`、`qwen-image-edit-2511-multiple-angles-lora.safetensors`、`qwen_2.5_vl_7b_fp8_scaled.safetensors`、`qwen_image_vae.safetensors` | `diffusion_models/`、`loras/`、`loras/`、`text_encoders/`、`vae/` | | **Story 4 / Storyboard 25** | `qwen_image_edit_2509_bf16.safetensors`、`Qwen-Image-Edit-2509-Lightning-4steps-V1.0-bf16.safetensors`、`next-scene-qwen-image-lora-2509.safetensors`、`qwen_2.5_vl_7b_fp8_scaled.safetensors`、`qwen_image_vae.safetensors` | `diffusion_models/`、`loras/`、`loras/`、`text_encoders/`、`vae/` | | **Multiangle**(3D 相机驱动) | 同上面 multi-view 套 | 同 | 下载: * Qwen Image Edit base:[https://huggingface.co/Comfy-Org/Qwen-Image-Edit\_ComfyUI](https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI) * Qwen Image base:[https://huggingface.co/Comfy-Org/Qwen-Image\_ComfyUI](https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI) * Lightning 4-step LoRA:[https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning](https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning) * Multiple-Angles LoRA:一般在 [https://huggingface.co/PVC-Inc/Qwen-Image-Edit-2511-Multiple-Angles-LoRA](https://huggingface.co/PVC-Inc/Qwen-Image-Edit-2511-Multiple-Angles-LoRA) * Next-Scene LoRA:[https://huggingface.co/lovis93/next-scene-qwen-image-lora-2509](https://huggingface.co/lovis93/next-scene-qwen-image-lora-2509) *** ## Relight | 工作流 | 文件 | 目录 | | ----------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------- | | **Flux2 Klein Relight** | `flux-2-klein-9b-nvfp4.safetensors`、`Sun_direction_LoRA_Flux_2_Klein_9b_v1.safetensors`、`qwen_3_8b_fp8mixed.safetensors`、`flux2-vae.safetensors` | `diffusion_models/`、`loras/`、`text_encoders/`、`vae/` | *** ## 全景(Panorama) | 工作流 | 文件 | 目录 | | ------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------- | | **Qwen-Image-Edit 2511 Image-to-Panorama**(图生全景) | `qwen_image_edit_2511_bf16.safetensors` *(或 fp8mixed)*、`Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors`、`qwen_2.5_vl_7b_fp8_scaled.safetensors`、`qwen_image_vae.safetensors` | `diffusion_models/`、`loras/`、`text_encoders/`、`vae/` | | **Qwen-Image 2512 + 360 LoRA**(文生全景) | `qwen_image_2512_fp8_e4m3fn.safetensors`、`Qwen-Image-2512-Lightning-4steps-V1.0-fp32.safetensors`、`qwen-360-diffusion-2512-int8-bf16-v2.safetensors`、`qwen_2.5_vl_7b_fp8_scaled.safetensors`、`qwen_image_vae.safetensors` | `diffusion_models/`、`loras/`、`loras/`、`text_encoders/`、`vae/` | 360 LoRA 下载:[https://huggingface.co/ProGamerGov/qwen-360-diffusion](https://huggingface.co/ProGamerGov/qwen-360-diffusion) *** ## 第三方 custom node 依赖 少数工作流依赖第三方 ComfyUI custom node: * **LaMa Erase** → [`Acly/comfyui-inpaint-nodes`](https://github.com/Acly/comfyui-inpaint-nodes) * **Fooocus SDXL Inpaint / Outpaint** → 用的 `INPAINT_*` 节点也由`comfyui-inpaint-nodes` 提供。 * \*(规划中,未提供)\***Demucs Vocals / Background** →[`lum3on/ComfyUI_AudioTools`](https://github.com/lum3on/ComfyUI_AudioTools)提供 `AudioStemSeparate` 节点。 其它工作流只用 ComfyUI 核心节点。 *** ## 不想用自带工作流? 不必下上面任何东西。ComfyTV 能跑任意 ComfyUI GUI 工作流 JSON,只要放进 `workflows//` 下面,再写一份 `_preset.json` 把 stage 的输入映射到工作流节点上即可。在常规 ComfyUI 里搭好工作流、保存、放进来、重启 ComfyUI,对应 stage 的工作流下拉里就会出现这个新选项。 `_preset.json` 的格式和实例见 [custom-workflows.zh.md](/zh/guides/custom-workflows)。