GPT Image 2.5 Sunburst: Image Edit
High-fidelity image editing template. Preserve original character pose, scene, lighting and composition. Only modify partial elements such as outfit, hairstyle, facial features or camera angle. Great for sequential frame generation with strong subject consistency.
简介
This ComfyUI template demonstrates a high‑fidelity image editing pipeline that preserves the original subject, pose, lighting, and composition while making targeted changes. It’s built around a simple, reliable graph: LoadImage feeds a reference image into OpenAIGPTImageNodeV2 for natural‑language edits, and SaveImageAdvanced handles consistent output naming and metadata. The included sample asset (hairstyle_makeup_base_gpt_image25.png) lets you test the flow immediately.
你需要提供
- 正在读取工作流
在 ComfyUI 中运行。 此工作流无法在本页面运行。你可以在 Comfy Cloud 中打开（已预先加载所需内容），或下载后在自己的 ComfyUI 中使用。
在 Cloud 中试用常见问答
No mask is required for this template. OpenAIGPTImageNodeV2 performs semantic, instruction‑based edits that aim to preserve the rest of the image. If you need pixel‑precise masked edits, consider an inpainting workflow that includes mask nodes.
Be explicit in your prompt about what must remain unchanged (e.g., “Keep the same pose, background, lighting, and framing”). Keep requested changes small and focused. Reusing the same base image and consistent phrasing across runs helps maintain fidelity.
Start with a moderate resolution that the API handles reliably (for example, around 1024 on the long side). Extremely large images may be resized by the API. If the node allows selecting output size, pick one that matches your use case and hardware/API limits.
Text‑guided editing can introduce variation. Keep prompts specific, avoid contradictory instructions, and change only one variable at a time. If the node supports multiple variations, compare them and iterate toward your target look.