Wan 2.7 Reference to Video: High-Fidelity Reference-Based Video Generation on playground and API | RunComfy

What makes Wan 2.7 Reference to Video stand out

Wan 2.7 Reference to Video is built for controlled video generation from image, video, or audio-guided references, with emphasis on subject fidelity, motion continuity, and scene consistency. This reference-to-video task converts reference assets into new video outputs that preserve identity and composition while following explicit motion and scene instructions. Wan 2.7 Reference to Video is suited to character-led clips, branded localization, and instruction-based sequence creation where stable visual carryover matters.

Key capabilities:

• Supports reference image URLs for appearance control across one or multiple subjects.
• Supports reference video URLs for both appearance and motion conditioning.
• Generates HD outputs at 720p or 1080p with aspect ratio control.
• Preserves subject identity and overall scene structure across frames.
• Handles short-form durations from 2 to 10 seconds for precise clip design.
• Offers single-shot or intelligent multi-shot generation workflows.
• Includes negative prompting and seed control for tighter repeatability in Wan 2.7 Reference to Video.

Prompting guide for Wan 2.7 Reference to Video

Start Wan 2.7 Reference to Video by supplying a clear prompt plus either reference images, reference videos, or both, depending on whether you need appearance transfer, motion transfer, or multi-subject consistency. Describe the subject, action, camera behavior, environment, and what must remain unchanged. For Wan 2.7 Reference to Video, keep instructions concrete: define motion pacing, framing, shot continuity, and visual constraints. Use negative_prompt to suppress unwanted traits, choose the aspect ratio based on delivery format, and enable multi_shots only when the sequence should break into coordinated cuts instead of one continuous take.

• Single character from images: "Use the reference images to keep the same person, generate a 5-second walking shot, medium framing, natural daylight, subtle camera follow."
• Motion-guided generation: "Use the reference video for movement style, preserve the subject appearance, create a 16:9 1080p clip with smooth forward motion."
• Multi-subject scene: "Use multiple reference images, keep each character distinct, stage them in one room, slow conversational gestures, stable composition."
• Brand localization: "Preserve product shape and colors from references, place in a modern retail environment, clean lighting, minimal camera drift."
• Instruction-based edit style output: "Match the reference subject, turn the scene into a rainy night street, cinematic reflections, no costume change."

Pro tips:

• In Wan 2.7 Reference to Video, separate appearance instructions from motion instructions.
• Use high-quality references with minimal occlusion and clear subject visibility.
• State what must stay fixed: identity, wardrobe, palette, layout, or camera distance.
• Keep durations aligned with the action complexity; shorter clips improve control.
• For more reference-to-video workflows, use the Wan 2.7 Reference to Video playground.

Note: If you need to modify an existing image, such as changing the background, lighting, or specific objects within a picture, use the Seedream 4.5 Edit model, which is optimized for instruction-based image manipulation.