AI Prompt for Video Generation (Sora, Veo, Runway)
A video-to-video restyle prompt for CogVideoX-5B that converts footage of abandoned Victorian mansion into anime Studio Ghibli 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 CogVideoX-5B. 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 abandoned Victorian mansion, currently shot in naturalistic style - **Target style:** anime Studio Ghibli - **Lighting target:** candlelit warm flicker - **Aspect:** 1:1 square - **Duration:** 5 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 CogVideoX-5B 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 anime Studio Ghibli) 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** (candlelit warm flicker) 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 ## CogVideoX-5B-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 anime Studio Ghibli 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 CogVideoX-5B offers it **C. Risk Callouts:** - Most likely artifact: [flicker / identity drift / background swim] - Mitigation: one concrete fix ## Strict Rules - No negative prompts unless CogVideoX-5B 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.