In traditional AI video generation, editors and animators constantly battle stochastic drift: single-image animation knows where it begins but rarely where it lands, while text-to-video hallucinates unexpected paths. Vidu's Start and End Frame mode solves this fundamental production bottleneck by establishing two hard visual boundaries—locking Frame 0:00 to an initial keyframe and the final frame to a target keyframe. By treating the AI generation engine as a physics-aware in-betweening and interpolation solver, creators can orchestrate precise match cuts, seamless morphs, and camera-guided transitions that fit neatly into commercial editing timelines.
Whether you are crafting narrative animations, high-impact commercial visual effects, or stylized motion graphics, here is how professional animators and video editors leverage Vidu's dual-frame workflow to maintain total creative control.
1. The Core Architecture: Why Start and End Frames Matter
Standard image-to-video tools operate like an open-ended camera pan: the model generates forward in time with decaying adherence to the initial prompt. In contrast, Vidu’s Start and End Frame mode functions like classical animation keyframing:
- Dual Visual Anchors: The diffusion model (powered by ShengShu Technology's U-ViT architecture) is conditioned simultaneously on the start image (
start_image_url) and the end image (end_image_url). - Non-Linear Interpolation: Rather than simple digital crossfading or optical flow morphing, Vidu computes 3D-aware depth, volume, lighting, and physical deformation across the sequence.
- Timeline Predictability: Because the final frame is predetermined, video editors can storyboard a sequence backward and forward, ensuring the outgoing tail of Shot A aligns perfectly with the incoming head of Shot B.
2. Keyframe Preparation: Setting Up the Assets
The success of a Vidu transition depends heavily on the visual hygiene of your two input keyframes. Feed the model conflicting canvas properties, and the interpolation will warp, stretch, or tear.
Aspect Ratio and Resolution Alignment
Both the start and end images must share identical aspect ratios (such as 16:9 for cinematic widescreen, 9:16 for vertical reels, or 1:1 for social feeds). If one frame is 1920×1080 and the other is slightly cropped to 1800×1080, Vidu must dynamically compensate for the aspect delta, resulting in unsightly edge stretching or floating geometry. Always standardize canvas boundaries in Photoshop, Figma, or your digital painting software before importing.
Spatial and Silhouette Anchoring
Vidu excels when there is a compositional throughline connecting the two frames:
- Focal Point Alignment: If your subject sits in the lower-third right of the opening keyframe, having the subject land in the same quadrant in the ending keyframe prevents violent spatial snapping.
- Silhouette Overlap: In transformational morphs (e.g., an organic seed blooming into a perfume bottle), maintaining a roughly similar silhouette or mass scale provides the model with a clear structural bridge.
- Lighting and Color Continuity: While Vidu can transition between day and night, drastic, unmotivated shifts in key light angle can cause mid-clip shadow flickering. Matching contrast curves and primary light sources yields substantially cleaner renders.
3. Step-by-Step Workflow: Web Studio and API
Vidu provides two primary entry points: the browser-based creation interface on the official web platform and developer API endpoints for programmatic pipelines.
Web Studio Execution
- Access Image-to-Video: Log in to Vidu and navigate to the creation workspace (
Image to Video). - Select Dual Keyframe Mode: In the media upload module, toggle the interface to enable both the First Frame (Start) and Last Frame (End) upload slots.
- Upload Keyframes: Drag and drop your high-resolution starting image into the first slot and your ending image into the second slot.
- Configure Generation Parameters:
- Duration: Choose between standard short clips (typically 4 seconds) or extended transitions (8 seconds) depending on the required pacing.
- Movement Amplitude / Motion Degree: For delicate facial shifts or subtle camera pushes, dial back motion amplitude to prevent unwanted warping. For high-energy morphs, kinetic camera moves, or visual effects, increase motion amplitude.
- Input the Kinetic Prompt: Detail how the transformation unfolds over time (see Prompting Strategies below).
- Generate and Review: Submit the job and review the rendered in-between motion.
Programmatic API Execution
For studios integrating automated storyboard rendering or ComfyUI pipelines, Vidu exposes dedicated endpoints via the Vidu API Platform:
- Endpoint:
POST /v1/vidu/start-end2video(or versioned platform route). - Payload Structure:
{ "prompt": "Smooth cinematic morph from sketch wireframe into photorealistic sports car, dynamic lighting, push-in camera, high fidelity", "start_image_url": "https://assets.yourstudio.com/frames/shot01_start.png", "end_image_url": "https://assets.yourstudio.com/frames/shot01_end.png", "duration": 4, "aspect_ratio": "16:9", "movement_amplitude": "auto" } - Poll task generation via the creation ID to download the completed MP4 sequence.
4. Prompt Engineering for Start-End Transitions: Describe the Bridge
The most common mistake made by new creators is writing prompts that re-describe Image A and Image B:
❌ Bad Prompt: "A sketch of a sports car on the left and a red sports car driving on a road at sunset on the right."
Because Vidu already extracts semantic tokens from both images, redundant visual descriptions confuse the model. Instead, describe the kinetic bridge—the mechanism, physics, and camera choreography required to travel from Point A to Point B.
The Kinetic Transition Formula
Structure your transition prompt using four key components:
[Mechanism of Change] + [Subject Interaction/Physics] + [Camera Trajectory] + [Temporal Quality & Stability]
Practical Prompt Archetypes
- Organic Morph / Metamorphosis:
"A continuous organic morph from the clay bust into a glowing bronze sculpture. Liquid metal flows seamlessly over the facial contours, cooling and hardening into polished metallic textures. Fixed locked camera, subtle specular reflections, smooth temporal coherence."
- Cinematic Whip-Pan & Match Cut:
"A rapid 90-degree whip-pan transition from left to right. The wooden corridor blurs into dynamic motion streaks before snapping crisply into the futuristic glass hallway. Fast camera motion blur, smooth acceleration and deceleration, seamless environmental handoff."
- Dolly-In Scale Transition:
"A steady, continuous camera push-in through the iris of the eye. The pupil expands into an abstract cosmic nebula, swirling with starlight and deep blue interstellar gas. Seamless zoom-through, zero geometric jitter, cinematic volumetric depth."
5. High-Impact Use Cases for Editors and Animators
1. Classical Pose-to-Pose Animation In-Betweening
Hand-drawn 2D and stylized 3D animators often spend days animating complex in-betweens (cushioning, follow-through, clothing drape). By painting or posing extreme Keyframe 1 (e.g., character preparing to leap) and Keyframe 2 (character landing), Vidu computes the secondary physics—flowing hair, fabric ripples, and weight shifts—saving dozens of production hours while keeping character designs locked.
2. Commercial Product Reveals and Exploded Assemblies
In commercial advertising, showing a packaged box transform into an assembled gadget or a transparent cross-section is an expensive CGI task. By rendering the unopened packaging as the start frame and the exploded mechanical assembly as the end frame, editors can use Vidu to generate seamless, high-tech unboxing shots.
3. Day-to-Night and Seasonal Time-Lapses
Architectural and documentary filmmakers frequently require time-lapse shifts over static structures. Setting a daytime exterior as Frame 1 and a twilight, window-illuminated facade as Frame 2 creates a photorealistic lighting progression complete with moving cloud shadows and sunset color gradients.
6. Pro Troubleshooting and Editorial Finishing
Even with state-of-the-art models, complex transformations can occasionally introduce visual artifacts. Here is how professional editors handle common edge cases:
- Mid-Sequence Geometric Collapses: If the semantic gap between Frame A and Frame B is too extreme (e.g., jumping from a medieval castle directly to a microscopic cell), the AI may generate messy artifacts around second 2.
- Fix: Implement a three-stage keyframe pass. Introduce a neutral intermediate bridge image (e.g., an abstract architectural drawing). Render clip 1 (Castle $\to$ Bridge) and clip 2 (Bridge $\to$ Cell), then splice them on the cut point in your NLE.
- Micro-Flicker Near Keyframe Landings: If the background lighting slightly disagrees between frames, the model may flutter just before hitting the final frame.
- Fix: Pre-grade the two images in DaVinci Resolve or Photoshop to equalize background luminance, black levels, and white balance prior to running the generation.
- Timeline Speed-Ramping (In NLE): Raw AI transitions often follow a linear temporal pace. In Premiere Pro, Final Cut Pro, or DaVinci Resolve, apply a gentle Bezier speed curve (slow start, rapid mid-transition snap, smooth landing) to elevate the cinematic polish. Complement the visual acceleration with custom sound design (whooshes, risers, and impact hits) to mask microscopic AI imperfections.
7. Pricing and Platform Realities
Vidu operates on a credit-based consumption model:
- Plan Tiers: Users can access the platform via free introductory trial credits, monthly/annual tiered subscriptions (billed through Stripe), or on-demand top-up credit packs.
- Credit Consumption: Credits are billed per generation task, with credit requirements scaling based on duration (4-second vs. 8-second outputs) and output quality tiers. Subscription credits renew monthly, while purchased top-up credits carry extended validity (up to 2 years).
- Enterprise Scaling: High-volume animation studios and automation teams can obtain dedicated API throughput and concurrency quotas via the Vidu API platform.
By treating Vidu's Start and End Frame mode not as a random lottery, but as a deterministic interpolation tool governed by rigid keyframe hygiene and kinetic prompt engineering, digital creators can seamlessly integrate generative video into exacting commercial production pipelines.