Create Images Using Qwen Image Edit 2511

Qwen Image Edit 2511 multi-reference image editing. Three inputs, one output: your subject placed into a new scene with matching lighting and environment.

Upload a main subject image, a lighting reference, and an environment reference. Describe the placement in natural language. The model repositions the subject into the environment, applies the lighting from the reference, and keeps the original identity, outfit, and facial features locked. The Lightning LoRA cuts generation to 4 steps.

The default prompt ships with a full production example: seat a subject on a red velvet chair from a reference interior, preserve every detail of their appearance, adjust the pose for sitting, blend shadows for contact realism, and apply dramatic cinematic lighting from the third reference. That's the full range of what this workflow does.

How do you use Qwen Image Edit 2511 for multi-reference image editing?

Upload your subject image, a lighting reference, and an environment reference. Write a prompt describing the placement and what to preserve. Qwen reads all three images together and composites the subject into the new scene. The Lightning LoRA runs the edit in 4 steps. Identity, outfit, and fine details stay locked from the main image.

Most runs need three things before hitting Run: the main image, at least one reference image, and a prompt that specifies placement and preservation.

Main image (subject anchor) The image the model treats as ground truth for identity. Facial features, hairstyle, makeup, jewelry, outfit, body proportions, and fine details all lock to this image. What you don't describe changing, the model keeps from here.

Secondary reference (lighting and mood) Upload an image whose lighting, color grade, or visual atmosphere you want applied to the output. The model reads this for soft highlights, colorful reflections, and cinematic glow rather than altering the subject's appearance.

Environment reference (scene and spatial context) The background or setting you want the subject placed into. The model reads scale, perspective, and spatial depth from this image and positions the subject to fit physically into the scene. No floating, no cutout look.

Positive prompt The most important input. Qwen Image Edit 2511 follows long, detailed instructions accurately. The default prompt runs to eight instructions covering placement, pose, identity preservation, scale, background replacement, shadow integration, and lighting application. That level of specificity works and produces clean results.

Tips for prompts that land: Describe what to add or change, not what to keep. Keeping things unchanged is the model's default. State preservation explicitly: "Preserve the subject's identity, facial features, hairstyle, outfit, and body proportions exactly." Name positions: "seated on the chair," "standing in the foreground," "correct scale and perspective." For lighting: name the style. "Cinematic glow," "soft highlights," "colorful reflections" all work. For contact realism: "accurate depth and contact shadows" stops subjects from looking pasted in.

Negative prompt List what you want to avoid. The default negative prompt targets the most common failures: face change, identity drift, floating body, broken anatomy, extra fingers, cutout look, flat lighting. Copy and extend this list for your use case.

CFG (default: 1) Default is 1 via CFGNorm. The community has found that 2.5 produces better results even with the Lightning LoRA active. Try 2.5 as your starting point. Higher values increase prompt adherence but can over-process fine detail.

LoRA strength (default: 1) The Lightning LoRA is loaded at full strength by default. This gives the fastest generation with clean results. For maximum character consistency where the Lightning LoRA is shifting the subject's look, the community recommends switching to the V1.0-fp32 variant over the V2.0 variant. V2.0 produces smoother, cleaner outputs but changes character more. V1.0 stays tighter to the original identity.

Steps (default: 4) 4 steps with the Lightning LoRA. This is enough for production-quality output. Increasing steps without adjusting the LoRA will produce diminishing returns.

Megapixels (default: 1.6MP) Output resolution. 1.6MP is set for sharp editorial-quality outputs. Lower it for faster preview runs.

What is Qwen Image Edit 2511 good for?

Qwen Image Edit 2511 is strongest for editorial composite work: placing a subject into a new environment while keeping their identity, outfit, and fine details intact. Three-reference architecture handles subject anchoring, lighting application, and scene placement in a single pass. Fastest path to photorealistic editorial composites without manual compositing.

Editorial portrait compositing. Place a subject from a studio shot into a different interior, outdoor scene, or set piece. The model preserves high-detail fashion elements like fabric, embroidery, and silhouette while adapting the subject's pose to fit the new environment. Contact shadows and depth integration are handled by the prompt.

Fashion and lookbook production. Reuse a single subject shoot across multiple backgrounds and lighting styles. Upload the same subject image, swap environment and lighting references, and generate a full set of editorial frames. Each output keeps the outfit and identity consistent while the scene and mood change.

Scene lighting replacement. Feed a well-lit reference image and describe the lighting transformation. Dramatic studio lighting, cinematic color grading, or environmental light changes all work. The model applies the lighting character from the reference without altering the subject's features.

Character placement in concept or production art. Place a character into a set reference for storyboards, pre-visualization, or concept reviews. The model handles perspective and scale matching between subject and environment, producing a grounded composite from a text description rather than manual masking.

Honest notes: at high CFG the model can over-process fine detail and shift identity. The V2.0 Lightning LoRA produces cleaner, smoother results but drifts from the original character more than V1.0. If character consistency is the priority, use V1.0-fp32. The TorchCompileModelQwenImage node can throw errors on some setups. Disable it if you encounter issues on first run.

How does Qwen Image Edit 2511 compare to single-reference image editors?

Qwen Image Edit 2511's three-reference architecture separates subject anchoring, lighting, and environment into distinct inputs. Single-reference editors apply all changes from one image and one prompt, which limits how much you can control each element independently. For composites where lighting and scene must come from separate sources, three references give you that control.

Single-reference editors (standard Klein 4B, SDXL inpainting) handle one anchor and describe everything else in the prompt. That works for simple edits: outfit swaps, background replacements, lighting adjustments. When you need the lighting from one image, the scene from a second, and the identity from a third, a single reference forces you to describe two of those three in text, which reduces precision.

The tradeoff: three inputs is more setup. For a simple background replacement or single-element edit, a single-reference workflow is faster. Qwen Image Edit 2511 earns its setup cost on editorial composites where precise control over all three elements produces outputs that would otherwise require manual compositing.

FAQ

How many reference images does Qwen Image Edit 2511 support?
This workflow uses three: a main subject image, a lighting reference, and an environment reference. Each serves a different purpose. The main image anchors identity and fine details. The lighting reference shapes mood and color. The environment reference defines the scene and placement context.

What CFG value works best with Qwen Image Edit 2511 Lightning LoRA?
The default is 1, but the community has found 2.5 produces better results even with Lightning active. Start at 2.5. Higher values increase prompt adherence but can shift identity at the edges. If faces are drifting, bring CFG back down toward 1.

How do I keep character identity consistent in Qwen Image Edit 2511?
Use the V1.0-fp32 Lightning LoRA variant rather than V2.0. V2.0 produces smoother, cleaner outputs but changes the character more. V1.0 stays tighter to the original identity. Also state preservation explicitly in your positive prompt: "Preserve the subject's identity, facial features, hairstyle, outfit, and body proportions exactly."

What should I put in the negative prompt for Qwen Image Edit 2511?
The default negative prompt covers the most common failures: face change, identity drift, floating body, broken anatomy, extra fingers, cutout look, pasted subject, flat lighting, harsh shadows. Copy this as your base and add anything specific to your subject or scene.

How do I run Qwen Image Edit 2511 online?
You can run Qwen Image Edit 2511 online through Floyo. No installation, no setup. Open the workflow in your browser, upload your reference images, and hit run. Free to try.