FireRed Image Edit ComfyUI Setup Guide
Set up FireRed Image Edit 1.1 in ComfyUI with the official workflow, correct model folders, multi-image inputs, and optional Lightning LoRA.
This FireRed Image Edit ComfyUI guide follows the workflow published by the FireRed Team. The official package includes a downloadable JSON workflow for FireRed-Image-Edit-1.1 and supports both single-image and multi-image editing.
What you need
Use a current ComfyUI installation and download the workflow from the official FireRed-Image-Edit-1.1-ComfyUI repository. You can also download the official firered-image-edit-1.1.json workflow directly. The workflow itself records the expected filenames and target directories, which is safer than following an old screenshot or third-party bundle.
The official workflow currently references these components:
| Component | Example filename | ComfyUI folder |
|---|---|---|
| Transformer | FireRed-Image-Edit-1.1-transformer.safetensors | models/diffusion_models |
| Text encoder | qwen2.5vl-7b-bf16.safetensors | models/text_encoders |
| VAE | qwen_image_vae.safetensors | models/vae |
| Optional Lightning adapter | FireRed Lightning 8-step .safetensors | models/loras |
The transformer is the largest file. Read FireRed Image Edit FP8 vs GGUF before choosing a quantized alternative.
FireRed Image Edit 1.0 vs 1.1
The official project describes 1.1 as an update built on the 1.0 foundation rather than a separate architecture. Choose the revision deliberately because the transformer, Lightning adapter, and workflow must agree.
| Revision | Official positioning | Practical reason to choose it |
|---|---|---|
| 1.0 | General-purpose instruction-based image editing | Reproduce an older workflow or compare against the original release |
| 1.1 | Improved portrait consistency, multi-element fusion, stylized text reference, and portrait makeup | Use the current official workflow for portrait, typography, and multi-reference work |
Do not load a 1.0 Lightning LoRA into a 1.1 graph simply because ComfyUI accepts the filename. The official package publishes revision-specific 1.0 and 1.1 adapters.
Official workflow reference
These node screenshots are the single-image and multi-image reference graphs linked from the FireRed Team's 1.1 ComfyUI package. Use them to check the overall node order; use the downloadable JSON above for exact settings.
Import and verify the workflow
- Update ComfyUI before importing the workflow. The official graph uses Qwen Image Edit nodes such as
TextEncodeQwenImageEditPlus. - Put each model file in the directory recorded by the workflow, then restart or refresh ComfyUI so the loaders can discover it.
- Drag the official
firered-image-edit-1.1.jsonworkflow onto the canvas. - Confirm that the transformer, text encoder, and VAE loader nodes resolve to real files instead of showing missing-model warnings.
- Load one source image, enter a direct editing instruction, and run the unmodified workflow once before adding custom nodes or LoRAs.
The official graph exposes up to three image inputs through the Qwen Image Edit conditioning node. Start with one clear source image; add image 2 and image 3 only when your prompt explicitly assigns a role to each reference.
Prompt structure for reliable edits
State what must remain unchanged before describing the edit. For example:
Keep the subject's face, pose, and camera angle unchanged. Replace the background with a clean beige studio, preserve realistic hair edges, and match the original light direction.
For multiple references, name them in order: “Use the person from image 1, the clothing from image 2, and the lighting from image 3.” This reduces ambiguity and makes failed inputs easier to diagnose.
Lightning LoRA and sampling
The official workflow contains an optional Lightning LoRA path. Its purpose is accelerated eight-step inference; it is not simply a universal quality upgrade. First verify the base workflow. Then enable the adapter and use the sampling values provided by the matching official workflow or model revision.
Do not mix a Lightning adapter from a different base version just because the filename loads. A graph can execute while producing unstable identity, weak edits, or unexpected texture if the adapter and transformer are incompatible.
Common failures
- A loader is red or empty: the file is in the wrong subdirectory, the filename changed, or ComfyUI has not refreshed its model list.
- The workflow reports missing node types: update ComfyUI and compare the workflow revision with your installation before installing arbitrary custom nodes.
- The output ignores image 2 or image 3: simplify the prompt and explicitly map each numbered image to one role.
- Out-of-memory errors: choose a lower-memory model format, reduce working resolution, close other GPU workloads, and test with one input image first.
For the canonical model description, license, and latest downloads, use the FireRed Team model card.
FireRed Image Edit Documentation
Independent technical guides for FireRed Image Edit 1.1, including ComfyUI setup, model format selection, and LoRA usage.
FireRed Image Edit FP8 vs GGUF vs BF16
Compare FireRed Image Edit 1.1 FP8, GGUF, and BF16 files by source, memory use, compatibility, and quality tradeoffs before downloading.