hi, what if i want to make videos mix realality and ai generated scenes? how to use seedance and it's alternatives, btw. i have no experiences with seedance. --- Yes—you can combine real footage with AI scenes very effectively. For a beginner, I recommend keeping the real teacher intact and using AI for the environment, cutaway shots, objects, or transitions. This preserves the teacher’s face, speech, and body motion. ## Three useful approaches ### 1. Real teacher over an AI-generated environment This is the most controllable workflow: ```text Film teacher normally ↓ Remove background with Kdenlive SAM2 ↓ Generate an empty AI environment ↓ Place teacher over environment in Kdenlive ↓ Add shadows, color matching and sound ``` For example, the teacher could appear inside: * A futuristic classroom * Ancient Rome * A space station * A microscopic cell * An animated mathematical world Generate the background without the teacher. A useful prompt is: > Empty futuristic classroom interior, presentation area in the center, cinematic but realistic, soft daylight from the left, locked camera, no people, no text, 16:9. “Locked camera” is important because an unnecessarily moving background makes compositing harder. ### 2. AI transforms part of the real shot Upload a short real clip and ask the model to replace or transform the environment: > Replace the classroom with the interior of a realistic orbital space station. Preserve the teacher exactly: same face, clothing, body, speech timing, gestures, framing and camera movement. This is faster, but the model may change the teacher’s face, fingers, clothing or lip movements. Use short clips and inspect every result. ### 3. Alternate between real and generated shots This often looks the most professional: ```text Teacher explains volcanoes → AI aerial shot of volcano → teacher continues → AI cross-section inside volcano → teacher concludes ``` You avoid trying to make one AI-generated shot last several minutes. AI video works best as short, purposeful shots edited together in Kdenlive. ## How to start with Seedance Seedance is ByteDance’s video-generation model. The official beginner-facing interface is **Dreamina**, while ByteDance also offers API-oriented access through BytePlus. Current Seedance versions accept combinations of text, images, video and audio references. [Official Dreamina Seedance page](https://dreamina.capcut.com/tools/seedance-2-0) A beginner workflow: 1. Open Dreamina and choose AI Video. 2. Start with **image-to-video**, not pure text-to-video. 3. Create or upload a reference image showing the desired scene. 4. Select landscape `16:9` for ordinary YouTube or educational videos. 5. Generate only a short shot—around 5–10 seconds initially. 6. Describe movement, camera, lighting and restrictions. 7. Download several variations. 8. Import the best result into Kdenlive. 9. Composite your isolated teacher above it. Image-to-video generally gives you more control because the initial composition, colors and object positions have already been decided. ### Seedance prompt structure Use this order: ```text Subject and action Environment Camera behavior Lighting and visual style Timing Things that must remain unchanged Things that must not appear ``` Example: > An empty orbital classroom with Earth visible through large windows. Computer displays glow subtly and stars move very slowly outside. Locked medium-wide camera with no zoom or panning. Realistic cinematic lighting, daylight entering from the left. Ten-second continuous shot. Keep the center floor area unobstructed for compositing a standing teacher. No people, no writing, no logos, no camera shake. For a video transformation: > Replace only the room with a realistic ancient Roman forum. Preserve the original teacher, face, hair, clothing, gestures, body position, timing and camera framing. Sunlight comes from the same direction as the original footage. Do not add other people near the teacher. Do not alter the mouth or hands. Reference material is especially important. Seedance can use visual references to help maintain characters, locations and styles; its earlier official model also supported text-to-video, image-to-video and multi-shot generation. [ByteDance Seedance overview](https://seed.bytedance.com/en/seedance) ## Seedance alternatives | Tool | Best use | Main tradeoff | | ------------------------ | -------------------------------------------------------------------------- | ------------------------------------------------------- | | **Seedance** | Cinematic generated scenes, multimodal references and shot sequences | Cloud service; results still need editing and selection | | **Runway Aleph** | Modifying existing real footage and replacing backgrounds or objects | Can unintentionally modify the person | | **Google Veo / Flow** | High-quality generated scenes, reference-based characters and environments | Access, quotas and cost vary | | **Blender** | Precise camera tracking, compositing, shadows and CG integration | More manual work | | **Local ComfyUI models** | Privacy, experimentation and automation on Ubuntu | Requires a powerful GPU and considerable setup | For your exact goal, **Runway Aleph is the strongest alternative when you want to transform footage you already recorded**. It is designed to replace backgrounds, objects, lighting and other elements in real video. [Runway Edit Studio](https://help.runwayml.com/hc/en-us/articles/51683104370451-Creating-with-Edit-Studio) **Veo** is particularly useful for generating separate cinematic scenes from reference images. It supports visual “ingredients” such as a character, object or environment reference. [Google DeepMind Veo](https://deepmind.google/models/veo/) ## My recommendation for your first project Make a 20–30 second experiment: 1. Record the teacher against an ordinary but uncluttered background. 2. Use Kdenlive SAM2 to isolate the teacher. 3. Generate a 10-second empty environment with Seedance. 4. Put the environment on Kdenlive track V1. 5. Put the transparent teacher on V2. 6. Add a soft contact shadow behind or beneath the teacher. 7. Adjust the foreground color to match the AI background. 8. Keep the original recorded voice instead of AI-generating it. Do not begin by asking Seedance to recreate the complete talking teacher. Speech, identity and hands are exactly the elements most likely to drift. Use AI as the **virtual set and B-roll generator**, while Kdenlive, Blender and FFmpeg remain your production backbone.