AI Prompt for Video Generation (Sora, Veo, Runway)
A video-to-video restyle prompt for Hunyuan Video that converts footage of child jumping in rain puddles into black and white noir while preserving motion.
More prompts for Video Generation (Sora, Veo, Runway).
An image-to-video prompt for Luma Dream Machine that animates child jumping in rain puddles using slow push-in over 15 seconds while preserving identity.
A multi-shot cinematic breakdown prompt that gives Lightricks LTX a full scene of coffee being poured into a ceramic cup in IMAX nature documentary with Instagram Reel intent.
An image-to-video prompt for Hailuo MiniMax that animates abandoned Victorian mansion using crane up over 15 seconds while preserving identity.
An image-to-video prompt for Kling 1.5 that animates lone tree on a windswept cliff using dolly right over 30 seconds while preserving identity.
Ready-to-paste Kling 1.6 prompt for a 5 seconds insert shot of underwater coral reef teeming with tropical fish rendered in hyperrealistic photorealism with gimbal smooth follow camera.
A multi-shot cinematic breakdown prompt that gives LTX Video a full scene of time-lapse of a blooming flower in cinematic 35mm film grain with game cinematic trailer intent.
You are writing a video-to-video (V2V) restyle prompt for Hunyuan Video. The user supplies source footage; the model re-renders it in a new visual style while preserving the original motion and composition. ## Inputs - **Source footage:** features child jumping in rain puddles, currently shot in naturalistic style - **Target style:** black and white noir - **Lighting target:** split lighting dramatic - **Aspect:** 1:1 square - **Duration:** 30 seconds ## V2V Prompt Philosophy V2V is fundamentally different from text-to-video: - The **motion, timing, and composition are LOCKED** by the source. - Your prompt controls **style, color, texture, and detail** — nothing else. - Describing motion is counterproductive; it fights the source and produces artifacts. - Describing the subject's identity strongly is critical, because Hunyuan Video may otherwise morph it frame-by-frame. ## Prompt Construction Write the prompt in this exact ordering: 1. **Subject anchor** (one clause: "a [subject description]") 2. **Style conversion** (2-3 style descriptors for black and white noir) 3. **Surface detail** (texture words: "hand-drawn line work", "oil paint impasto", "pixel dithering") 4. **Color palette** (3-5 specific colors or a named palette) 5. **Lighting match** (split lighting dramatic) 6. **Anti-flicker anchors** (phrases that reduce frame-to-frame inconsistency: "temporally coherent", "stable linework", "consistent character design") Do NOT include: - Camera motion descriptions - Subject action verbs - Environmental changes - Time-of-day shifts not present in the source ## Hunyuan Video-Specific V2V Notes - **Runway Gen-3 Alpha:** Use their Video-to-Video feature with "structure strength" around 0.6-0.8. Lower = more stylization, higher = more faithful to source. - **Pika 2.0:** "Modify Region" works well for localized restyles. For full-scene, keep structure strength high. - **Kaiber / Gen-2 legacy:** Use high CFG (12-15) with short prompts. - **Kling 1.6:** Has a dedicated Motion Brush — prompt the style, let the brush handle regional intensity. - **Luma Modify:** Allows style reference images; attach a black and white noir reference if possible. - **Open-source (AnimateDiff + ControlNet):** If the user is routing through ComfyUI, mention they should use ControlNet Tile + Temporal Net for coherence. ## Output Format **A. The Prompt** (code block, 50-90 words): ``` [One paragraph, 50-90 words, no motion language] ``` **B. Suggested Settings:** - Structure strength: [0.5-0.85 depending on how drastic the restyle is] - Denoise / CFG: [platform-specific] - Frame rate: match source - Consistency: enable if Hunyuan Video offers it **C. Risk Callouts:** - Most likely artifact: [flicker / identity drift / background swim] - Mitigation: one concrete fix ## Strict Rules - No negative prompts unless Hunyuan Video explicitly supports them (Runway and Kling do; Sora does not). - Keep the prompt under 90 words — long V2V prompts cause regional attention bleeding. - Do not describe audio (V2V is silent by default; Veo 3 is the exception). Generate the V2V prompt now.
Replace the bracketed placeholders with your own context before running the prompt:
[subject description]— fill in your specific subject description.[One paragraph, 50-90 words, no motion language]— fill in your specific one paragraph, 50-90 words, no motion language.[0.5-0.85 depending on how drastic the restyle is]— fill in your specific 0.5-0.85 depending on how drastic the restyle is.[platform-specific]— fill in your specific platform-specific.[flicker / identity drift / background swim]— fill in your specific flicker / identity drift / background swim.