Pick the right AI video tool for every shot
Text-to-video, image-to-video, start and end frames, lip sync: which method and which model for each kind of shot, how to test them in ten minutes, and how to spend your credits where they count.
Every shot has a job
Beginners pick one AI video tool and ask it to do everything. Experienced creators do the opposite: they look at what each shot must do, then choose the method and the model that does that job best.
A wide shot of a city must look grand and stable. A close-up must keep your hero's face exactly right. A dialogue shot needs lips that match the voice. A fight needs fast, readable motion. These are four different problems, and no single tool is the best at all four.
- Wide shot of a place
- Text-to-video or image-to-video
- Close-up of your hero
- Image-to-video from the character sheet
- Two characters talking
- A lip-sync tool
- Fast action
- Start and end frames
- Magic, fire, explosions
- Image-to-video, effects added in the edit
- Decide what the shot must do first. The tool comes second.
Four ways to make a shot
Almost every AI video tool offers the same four methods under different names. Knowing them is more useful than knowing any single app.
- Text-to-video: you describe the shot in words. Fast for places, crowds and moods, but the face of your hero will change from one clip to the next.
- Image-to-video: you give a still image and describe the motion. The best way to keep characters, costumes and light consistent. Most series are made this way.
- Start and end frames: you give the first and the last image, and the model invents the motion between them. Ideal for transformations, a punch that lands, a door that opens on a precise place.
- References (also called elements or ingredients): you give separate images of a character, an object and a place, and the model combines them in a new shot. Useful when you need the same hero in many new places.
Text-to-video for every shot of the episode: "a young archer with red hair walks in the forest". In each clip she has a different face, a different bow and a different coat.
One approved still of the archer from the character sheet, then image-to-video for each shot: "she walks slowly between the pines, looking up, wind in her hair". Same face, same bow, same coat.
When in doubt, start from a still image. A failed still costs a fraction of a failed video clip, and it gives the model a clear starting point.
What each family of models is good at
Models change every month, so treat this as a starting point to test, not a ranking. As of November 2026, this is what creators on DeepStories most often report.
- Kling: good value, strong character references and built-in lip sync. A favourite for series with recurring heroes.
- Seedance (in Dreamina and other platforms): cinematic camera work and scenes with several shots in one generation.
- Veo (in Google Flow): very realistic images and native sound, at a higher price per clip.
- Hailuo (MiniMax): cheap, energetic motion; good for action and tests.
- Runway: editing tools and control over an existing clip, more than raw generation.
- Wan: open models you can run on your own computer if you have a strong graphics card.
Pick one main model and learn its habits. You will get more usable shots from one model you know than from five you don't.
The ten-minute test: three models, one shot
Before you spend a month of credits on one tool, run a quick test on the hardest shot of your episode. Use the same still image and the same prompt in three models, then score each result.
- Motion: does it move naturally, without melting or jumping?
- Face: is your hero still your hero at the last frame?
- Prompt: did it do what you asked, and only that?
- Cost: how much did one usable clip cost, counting the failed tries?
- Model A
- Model B
- Model C
- Motion
- Face stays the same
- Follows the prompt
- Cost per usable shot
- Model B wins this shot
- Score each model from 1 to 5 on the same shot. The winner can change from one kind of shot to another.
Voices and lip sync
Dialogue is where AI films most often look fake. The fix is mostly in the method, not the tool.
- Record or generate the voice first, then animate the face to it.
- Keep talking shots short: 2 to 4 seconds, then cut to the listener or to a detail.
- Prefer medium close-ups facing the camera; profiles and fast head turns break lip sync.
- Let some lines play off-screen, over a reaction or a landscape. Real films do it all the time.
One platform or several?
You can subscribe directly to one model, or use a platform that gives access to several models with one account. Direct subscriptions are usually cheaper per clip. Multi-model platforms let you pick the best model for each shot without paying five subscriptions.
A common setup: one direct plan for the model you use for most shots, and a small amount of credits on a multi-model platform for the few shots it can't do well.
Worked example: one scene, tool by tool
An archer discovers her village burned at dawn. Six shots, and the method chosen for each.
- Wide shot of the burned village in the mist: image-to-video from a still made in your image tool, slow push-in.
- Close-up of the archer's face, eyes opening: image-to-video from the character sheet, almost no motion.
- She stands up among the ashes: image-to-video, medium shot, the motion described in one sentence.
- Insert, a royal arrow stuck in a beam: image-to-video, very small camera move. Inserts are cheap and hide cuts.
- She says "Who did this?": voice generated first, then a lip-sync pass on a 3-second medium close-up.
- She runs toward the city gate: start and end frames, so she arrives exactly at the gate you designed.
Notice that only one shot out of six needs lip sync, and only one needs start and end frames. Most shots are simple image-to-video, which keeps the budget low.
Let an AI assistant plan the tools with you
Paste your shot list into an AI assistant and ask it to suggest a method for each shot. You stay the director; it saves you time.
Here is the shot list of my AI series episode: [paste your shot list]. For each shot, tell me which method to use (text-to-video, image-to-video, start and end frames, references, or lip sync) and why, in one line. Mark the 3 hardest shots, the ones I should test in several models first. Keep my shot numbers.
Final checklist
- Each shot in your list has a method next to it.
- Every character shot starts from an approved still image.
- You tested the hardest shot in three models and kept the winner.
- Dialogue shots are short and face the camera.
- Most of your credits go to the shots viewers will remember.
- Decide what each shot must do, then pick the method and the model.
- Start character shots from an approved still image.
- Test your hardest shot in three models before committing your credits.
- Kling: Character references, lip sync, good value.
- Dreamina: Seedance models for cinematic shots.
- Google Flow: Veo, realistic shots with native sound.
- Hailuo: Cheap, energetic motion for tests and action.
- Runway: Editing and control over existing clips.
- ElevenLabs: Voices to drive your lip-sync shots.
Pick the right AI video tool for every shot — DeepStories Academy