[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-92295":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":18,"hasPages":18,"topics":20,"createdAt":9,"pushedAt":9,"updatedAt":21,"readmeContent":22,"aiSummary":23,"trendingCount":15,"starSnapshotCount":15,"syncStatus":14,"lastSyncTime":24,"discoverSource":25},92295,"ComfyUI-Krea2-Ostris-Edit","ostris\u002FComfyUI-Krea2-Ostris-Edit","ostris","Comfy UI Nodes for Krea 2 LoRAs trained with AI Toolkit Experimental Edit",null,"Python",129,1,52,2,0,58,43.7,false,"main",[],"2026-07-22 04:02:05","# ComfyUI-Krea2-Ostris-Edit\n\nNodes for running Krea 2 edit LoRAs trained with ai-toolkit (`krea2` arch with\n`model_kwargs.edit: true`). \n\n## Installation\n\nClone this repo into your ComfyUI `custom_nodes` folder and restart ComfyUI:\n\n```bash\ncd ComfyUI\u002Fcustom_nodes\ngit clone https:\u002F\u002Fgithub.com\u002Fostris\u002FComfyUI-Krea2-Ostris-Edit.git\n```\n\nNo extra dependencies are required. The nodes appear under the `ostris\u002Fkrea2`\ncategory.\n\n## Nodes\n\n### Text Encode Krea 2 Ostris Edit\n\nInputs: `clip`, `prompt`, optional `vae` and `image1`..`image3`.\nOutput: `CONDITIONING`.\n\nEncodes the prompt together with the reference images through the Krea 2\nQwen3-VL text encoder, using Krea's conditioning template with\n`Picture N:` vision placeholders — the same layout used during training.\nWhen a VAE is connected, each reference image is also VAE-encoded and attached\nto the conditioning as reference latents for the model patch node.\n\nImage sizing matches training: images fed to the Qwen3-VL encoder are\ndownscaled (never upscaled) to fit 384x384 total pixels; reference latents to\nfit 1MP.\n\nNote: the text encoder checkpoint must include the Qwen3-VL vision weights or\nthe images cannot be encoded.\n\n### Krea 2 Ostris Edit Model Patch\n\nInput: `model`. Output: `model`.\n\nPatches the Krea 2 model so it consumes the reference latents from the\nconditioning. Each reference is appended to the image token sequence and\nconditioned at timestep 0 (the `index_timestep_zero` reference method), and\nthe denoising prediction covers only the target image tokens. This node is\nrequired because the stock Krea 2 model in ComfyUI ignores reference latents.\n\nIf the conditioning has no reference latents, the patched model behaves\nexactly like the stock model, so it is safe to leave in the graph.\n\n## Example wiring\n\n```\nLoad Diffusion Model (krea2) -> Load LoRA -> Krea 2 Ostris Edit Model Patch -> KSampler\nCLIPLoader (krea2) -> Text Encode Krea 2 Ostris Edit (prompt + images + VAE) -> positive\nCLIPLoader (krea2) -> Text Encode Krea 2 Ostris Edit (negative prompt)      -> negative\n```\n","这是一个为ComfyUI设计的插件，用于支持Krea 2模型的编辑型LoRA（基于AI Toolkit实验性Edit训练流程），实现图文联合条件控制的图像生成。核心功能包括：支持多参考图输入的专用文本编码节点（集成Qwen3-VL多模态编码器，自动处理图像缩放与VAE编码）和模型补丁节点（使Krea 2模型能正确消费参考图像的潜变量并执行timestep-0索引式条件注入）。适用于需要以参考图引导风格\u002F结构编辑的AI图像生成场景，如局部重绘、多图融合、视觉一致性编辑等。","2026-07-08 04:30:00","CREATED_QUERY"]