How do I make an AI video with the same characters in every shot?
Design the character once as reference images, then feed those same references into every image and video generation instead of re-describing the character in text. On Gulab you do this on a node canvas: a character image (or an Element holding several pictures of it) is wired into each shot, and video models like Seedance 2.0, Kling V3 and Hailuo 03 animate from it. You can see exactly which references every shot used, so drift is easy to catch and fix.
Why characters drift in AI video
Text-to-video models invent a new person every time you describe one in words. "A woman in a red coat" produces a different face, build and coat on each run. The fix is to stop relying on the prompt for identity and give the model pictures: a locked face, outfit and silhouette that every shot starts from. Character consistency is a reference-management problem more than a prompting problem, which is why a canvas where references are visible objects helps.
The workflow in Gulab, step by step
- Create the character sheet. Add an Image node and generate the character on a clean background: front view, three-quarter view, full body. Image models on Gulab include Nano Banana Pro, Nano Banana 2, Seedream v4.5, Seedream 5.0 Pro, GPT Image 2.0, FLUX 2 Pro and FLUX 3. Or upload your own drawing or photo of an actor you have rights to.
- Save it as an Element. Elements hold several pictures of one subject and move between boards through a shared clipboard, so the same character can be reused across projects.
- Generate shots from the reference. Wire the character into new Image nodes for each key frame. Connecting a reference image to models like FLUX 3 or Ideogram 4.5 switches to their edit model automatically, and FLUX 3 Edit accepts up to 10 reference images. In the prompt, @-tag the reference ("@Ava walks into the station") and the tagged reference gets a red ring so you can see what the model is looking at.
- Animate each key frame. Connect the key frame to a Video node. Image-to-video keeps the character because the first frame already contains them. For multi-reference shots, Seedance 2.0 has a references mode, and Hailuo 03 and Happy Horse 1.1 offer reference-to-video.
- Give them a voice. Use Text to Speech, Voice Clone or Voice Recorder for dialogue, then Voice Sync (AI lip sync) to match mouth movement. Hailuo 03 Max Lip Sync turns a still photo plus 5 to 15 seconds of audio into a talking video.
- Cut it together. Concat, Crossfade, Audio Mix, Text Overlay and Export nodes assemble the final video on the same canvas.
Which video models to use for character shots
| Model (5-second clip) | Good for | Price |
|---|---|---|
| Seedance 2.0 (480p / 720p / 1080p) | Multi-reference shots, motion | $0.45 / $0.97 / $2.13 |
| Seedance 2.5 (480p / 720p) | Longer, more complex motion | $0.69 / $1.48 |
| Kling V3 Pro | Image-to-video with first and last frame | $1.01 |
| Kling O3 Standard | Cheaper image-to-video and text-to-video | $0.67 |
| MiniMax Hailuo 03 (2K + audio) | Reference-to-video with sound | $1.56 |
| Veo 3.1 Fast (720p) | Fast drafts | $0.90 |
Prices are the per-generation costs listed on gulab.ai/pricing in October 2026 and can change; the pricing page is always the live source.
Tips that keep a character on-model
- Keep the outfit and hairstyle in the reference, not only in the prompt. Prompts that contradict the reference cause drift.
- Generate key frames as images first and approve them, then animate. Re-running a cheap image is better than re-running an expensive video.
- For a shot that must start and end on specific poses, use a model with first and last frame support (Kling V3 Pro, Veo 3.1). Kling V3 Turbo does not support an end frame; Gulab tells you and switches to Kling V3 Pro for that run.
- Group shots with Heading and Zone nodes so a long project stays readable.
- Ask the canvas assistant to set it up for you. It can place nodes, wire references and pick models; one of its starter ideas is literally "one character in 5 animation styles".
Pricing
Gulab bills per generation from a credit balance. Plans are Solo Producer ($10/month, ₹799), AI Era Creator ($20/month, ₹1,599) and Full Studio ($50/month, ₹3,999). Every plan includes every model, with no model-gating. Unused credits from a monthly plan roll over for up to 60 days, and top-up credits (available to subscribers) last until used.
FAQ
Can I keep the same face across different AI video shots?
Yes. Use the same character reference image (or an Element with several pictures of the character) as input to every shot, generate key frames from it, then animate those frames with image-to-video. Identity comes from the picture, not the text.
Which AI video models support reference images on Gulab?
Seedance 2.0 has a references mode, Hailuo 03 and Happy Horse 1.1 offer reference-to-video, and every image-to-video model (Kling, Veo, Seedance, Wan, LTX, Grok Imagine and others) starts from the image you connect.
Can my character talk?
Yes. Generate or record a voice with Text to Speech, Voice Clone or Voice Recorder, then use Voice Sync for lip sync. Hailuo 03 Max Lip Sync turns a still photo plus 5 to 15 seconds of audio into a talking video.
How much does a character video cost?
Each generation is billed separately. A 5-second Seedance 2.0 clip at 480p is $0.45 and a Kling V3 Pro clip is $1.01, per gulab.ai/pricing in October 2026. Plans start at $10 a month and include every model.
Do I need to write prompts myself?
No. The canvas assistant can build the graph for you: character sheet, key frames, video nodes and the final cut, which you can then edit node by node.