logo
RunComfy
ComfyUIPlaygroundPricing
discord logo
ComfyUI>Workflows>Qwen Edit 2509 Light Restoration | Photo Relight Tool

Qwen Edit 2509 Light Restoration | Photo Relight Tool

Workflow Name: Qwen Edit 2509 Light Restoration
Workflow ID: 0000...1311
This workflow helps you achieve natural and even lighting across portraits or product shots. It removes harsh shadows and bright glare while maintaining subject detail and texture. Built for precise, tone-consistent image correction, it ensures your edits look subtle and professional. Perfect for optimizing lighting in photos taken under inconsistent conditions. You'll save editing time while gaining full control over light and tone balance for your visuals.

Qwen Edit 2509 Light Restoration: turn harsh lighting into soft, natural illumination

Qwen Edit 2509 Light Restoration is a focused ComfyUI workflow for quickly fixing difficult lighting. It evens out exposure, softens hard shadows, and restores neutral light while protecting subject identity and fine detail. Designed for portraits, product photos, and aesthetic cleanup, the workflow produces clean, professional results from challenging source images.

Built on Qwen’s image-editing stack with dedicated Light Restoration and Relight adapters, this graph pairs strong text-driven control with fast sampling. Use it when you need reliable, repeatable lighting correction without changing the look of the scene. Throughout this README, the term Qwen Edit 2509 Light Restoration refers to the complete workflow as provided.

Key models in Comfyui Qwen Edit 2509 Light Restoration workflow

  • Qwen-Image-Edit-2509 (UNet, GGUF). The base image-editing backbone that interprets lighting instructions and edits in latent space while preserving structure.
  • Qwen Image VAE. Handles latent encoding/decoding for the Qwen image domain. Source: Comfy-Org/Qwen-Image_ComfyUI.
  • Qwen2.5-VL 7B visual-language encoder. Provides multimodal text conditioning aligned to Qwen Image Edit 2509 for precise, instruction-style edits.
  • Qwen-Edit-Light_Restore LoRA. Specializes the edit model to remove harsh highlights, fill shadows, and normalize light. Source: dx8152/Qwen-Image-Edit-2509-Light_restoration.
  • Qwen-Edit-Relight LoRA. Enables targeted relighting with clean, directional light when you want to creatively re-illuminate the scene. Source: dx8152/Qwen-Image-Edit-2509-Relight.
  • Qwen-Image-Lightning-8steps V2.0 LoRA. A speed and stability adapter that makes high-quality edits possible in very few steps. Source: lightx2v/Qwen-Image-Lightning.

How to use Comfyui Qwen Edit 2509 Light Restoration workflow

The graph contains two independent paths you can run separately: Relight and Light Restore. Both start by resizing the image to a safe working resolution, encode to latents, apply Qwen Edit 2509 adapters, sample, decode, and then preview or save. Use the Compare viewers to verify the improvement before you export.

  • Image1

    • Purpose: input for the Relight path. Load your photo with LoadImage (#31). The workflow scales it automatically using ImageScaleToTotalPixels (#39) to balance quality and speed.
    • How it works: the scaled image is encoded to latents with VAEEncode (#10) so edits happen in a compact representation that preserves detail.
    • What you provide: a single RGB image. Use a well-exposed file; extremely blown highlights or fully black shadows leave little to recover.
  • Relight

    • Purpose: creative relighting that adds clear, neutral midday light with defined yet natural shadows. This is ideal when you want a new lighting direction rather than only softening flaws.
    • How it works: enter an instruction in TextEncodeQwenImageEditPlus (#11). The default prompt starts with “Relight:” which acts as a trigger for the relight adapter. The model fuses your instruction with the image latent and applies the Relight and Lightning adapters during sampling.
    • What you provide: keep the “Relight” trigger word and describe the lighting intent (for example “neutral color temperature, soft edge shadows”). The output is decoded with VAEDecode (#12), previewed, and shown side-by-side in Image Comparer (rgthree) (#44).
  • General (Relight)

    • Purpose: model assembly and sampling. The UNet is enhanced with the Relight adapter plus Lightning for fast, stable steps, then normalized with CFG before sampling.
    • How it works: the assembled model is routed through ModelSamplingAuraFlow and CFGNorm to stabilize guidance, then KSampler (#14) produces the edited latent.
    • What you provide: optional seed control for repeatability and any wording tweaks in the instruction. The rest is preconfigured for the Qwen Edit 2509 Light Restoration use case.
  • Picture1

    • Purpose: input for the Light Restore path. LoadImage (#93) brings in your file, and ImageScaleToTotalPixels (#84) prepares a pixel budget that avoids memory spikes while keeping detail.
    • How it works: the resized image is used both for a live before/after in Image Comparer (rgthree) (#94) and as the source for encoding and restoration.
    • What you provide: a single image in need of softening or exposure smoothing.
  • Light Restore

    • Purpose: the core Qwen Edit 2509 Light Restoration flow that removes harsh, uneven lighting and returns a clean, neutral look. It is designed for subtle, photoreal fixes.
    • How it works: enter an instruction in TextEncodeQwenImageEditPlus (#99) that asks to remove existing light artifacts and apply soft, even light. Negative conditioning TextEncodeQwenImageEditPlus (#92) can limit unwanted style shifts. The restored latent is decoded high-res with VAEDecodeTiled (#103) and exported via SaveImage (#96).
    • What you provide: a concise, plain-language instruction such as “soft, even lighting, neutral color, preserve subject details.” Use the viewer to confirm that texture and identity are intact.
  • General (Light Restore)

    • Purpose: model assembly and stabilized sampling tailored for restoration. The base Qwen-Image-Edit-2509 UNet is paired with the Light_Restore and Lightning adapters, then guided through a consistent scheduler and CFG normalization.
    • How it works: VAEEncode (#91) prepares latents, ModelSamplingAuraFlow (#87) sets a Qwen-friendly schedule, CFGNorm (#86) keeps guidance balanced, and KSampler (#83) performs the edit. VAEDecodeTiled (#103) ensures large outputs decode without seams.
    • What you provide: optional seed, plus any small text refinements to control softness and neutrality. Output can be compared against the original with a draggable slider in Image Comparer (rgthree) (#94).

Key nodes in Comfyui Qwen Edit 2509 Light Restoration workflow

  • TextEncodeQwenImageEditPlus (#11)

    • Role: turns your relight instruction into conditioning aligned with Qwen Edit 2509. Use imperative, concrete phrases and keep the “Relight” trigger at the start to ensure the adapter fully engages. Adjust wording to control direction, softness, and color temperature.
  • TextEncodeQwenImageEditPlus (#99)

    • Role: primary instruction encoder for the Light Restoration path. Keep prompts short and literal; emphasize “soft, even lighting” and “preserve details.” If the result looks stylized, add a short negative constraint in TextEncodeQwenImageEditPlus (#92).
  • LoraLoaderModelOnly (#98)

    • Role: applies the Light_Restore specialization that targets highlight rolloff and shadow lift. Increase the adapter weight for stronger correction; lower it for minimal, documentary changes. Pairing it with the Lightning adapter keeps edits crisp at low step counts.
  • ModelSamplingAuraFlow (#2)

    • Role: selects a scheduler tailored to Qwen’s editing dynamics. Use it to maintain stable guidance at low steps; switch only if you have a specific reason to trade smoothness for a different texture.
  • KSampler (#83)

    • Role: the main denoising engine. Tune seed for reproducibility, adjust steps to balance speed and fidelity, and nudge cfg if you need stronger adherence to the instruction or more preservation of the original scene.
  • VAEDecodeTiled (#103)

    • Role: decodes large images without seams. If you see borders or memory pressure, adjust tile and overlap to match your GPU budget; larger tiles reduce texture mismatches, smaller tiles reduce VRAM use.
  • ImageScaleToTotalPixels (#84)

    • Role: sets a total megapixel target so you can process large inputs safely. Raise the pixel budget for more detail when you have VRAM headroom; lower it when batch-processing many images quickly.

Optional extras

  • Keep instructions literal. Qwen Edit 2509 Light Restoration responds best to short, descriptive phrases rather than artistic prompts.
  • Use the “Relight” trigger word when running the creative relight path; leave it out for the Light Restore path.
  • If faces or logos must not change, add short negatives like “no identity change, no style change” in the negative field.
  • For products, mention surface qualities you want to preserve (for example “retain metal texture, avoid plastic look”) so the restoration protects micro-contrast.
  • Set a fixed seed to lock in a look, then sweep steps upward only if you notice residual banding or unevenness.

Acknowledgements

This workflow implements and builds upon the following works and resources. We gratefully acknowledge dx8152 for Qwen Edit 2509 Light Restoration and dx8152 for Qwen Edit 2509 Relight for their contributions and maintenance. For authoritative details, please refer to the original documentation and repositories linked below.

Resources

  • dx8152/Qwen Edit 2509 Light Restoration
    • Hugging Face: dx8152/Qwen-Image-Edit-2509-Light_restoration
  • dx8152/Qwen Edit 2509 Relight
    • Hugging Face: dx8152/Qwen-Image-Edit-2509-Relight

Note: Use of the referenced models, datasets, and code is subject to the respective licenses and terms provided by their authors and maintainers.

Want More ComfyUI Workflows?

SeedVR2 | Image & Video Upscaler

Fixes blur instantly. Better than Keep/PMRF.

Flux Consistent Characters | Input Image

Flux Consistent Characters | Input Image

Create consistent characters and ensure they look uniform using your images.

PMRF Ultra Fast Upscaler | Low VRAM ComfyUI

Ultra fast PMRF upscaler! 3.79s on medium machine. 2x scale.

Audioreactive Dancers Evolved

Transform your subject with an audioreactive background made of intricate geometries.

SUPIR | Photo-Realistic Image/Video Upscaler

SUPIR enables photo-realistic image restoration, works with SDXL model, and supports text-prompt enhancement.

Flux Kontext Pulid | Consistent Character Generation

Create consistent characters using FLUX Kontext with a single face reference image.

ACE++ Face Swap | Image Editing

Swap faces in images with natural language instructions while preserving style and context.

Mesh Graphormer ControlNet | Fix Hands

Mesh Graphormer ControlNet | Fix Hands

Mesh Graphormer ControlNet corrects malformed hands in images while preserving the rest.

Follow us
  • LinkedIn
  • Facebook
  • Instagram
  • Twitter
Support
  • Discord
  • Email
  • System Status
  • Affiliate
Resources
  • Free ComfyUI Online
  • ComfyUI Guides
  • RunComfy API
  • ComfyUI Tutorials
  • ComfyUI Nodes
  • Learn More
Legal
  • Terms of Service
  • Privacy Policy
  • Cookie Policy
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Playground, enabling artists to harness the latest AI tools to create incredible art.