Talk-to-edit conversation
Generate, then refine turn by turn. Characters and scene persist across every edit.
Google · Live on Coinis
Generate a clip. Then change it by talking. Native synced audio. Real physics. Live on Coinis today.
No credit card. 25 free tokens a week.
The fast tier beside Veo 3 · All Google models on Coinis
Why Omni Flash
Most AI video is one prompt, one gamble. Omni Flash is a conversation. Draft fast, then steer it change by change.
Generate, then refine turn by turn. Characters and scene persist across every edit.
Text, images, video and audio in one request. One model reads them all.
Sound is generated with the picture. Write SFX cues at exact seconds. They land.
Grounded in Gemini. Gravity, fluids and collisions behave like the real thing.
Showcase
We generated every clip ourselves with Gemini Omni Flash. Every prompt is shown exactly as run. Copy one and remix it in the app.
Talk-to-edit
This is a real conversation with Omni Flash. Turn one generates the base clip. Each next turn is one instruction that builds on the last. Same skater, same run, same camera.
Turn 1 · Base generation
Turn 2 · Talk-to-edit
Keep the same skateboarder, the same run and the same camera movement, but transform the parking garage into a rain-soaked futuristic neon alley at night, puddles reflecting pink and cyan light, light rain falling, wheels now hissing over wet asphalt. Every neon sign and light panel is a pure abstract glowing shape — arcs, rings, chevrons and bars of light — with absolutely no letters, no words, no characters and no writing of any language anywhere in the scene.
Turn 3 · Stacked edit
Now make the skateboard leave a ribbon of glowing golden light behind it that hangs in the air along the whole path, reflecting in the puddles, everything else unchanged.
Under the hood each turn passes the previous interaction id back to Google's Interactions API. The model remembers the scene, so you only describe the change.
Capabilities
Direct the frame second by second. Trust the physics. Go vertical for social. Each tab is a real generation with its prompt as run.
Write the timeline into the prompt. Omni Flash hits each beat. The native audio lands with it.
Omni Flash in the wild
Brand ads, animation, gaming, music video, documentary, product film. Filter the gallery. Copy any prompt.
Omni Flash vs Veo 3
Omni Flash is the fast conversational tier. Veo 3 is the high-fidelity tier. Both run on Coinis from the same token balance. Draft in one, finish in the other.
| Capability | Omni Flash New | Veo 3 |
|---|---|---|
| Clip length | Up to 10 seconds, longer coming | Longer clips |
| Resolution | 720p | Up to 4K |
| Editing | Talk-to-edit, multi-turn conversation | Prompt-based generation |
| Audio | Native synced audio | Native synced audio |
| Price per second | $0.10 on Google's API | Premium tier |
| Built for | Speed and iteration | Final-quality fidelity |
Veo 3 prompts, examples and specs live on its own page. Or browse every Google model on Coinis.
Prompt guide
Every prompt on this page follows the same grammar. Four rules cover most of it.
Name the shot type and the camera move. Wide, macro, handheld tracking, slow dolly in.
One main subject per prompt. Around 200 words is the stability sweet spot.
Write 0-3s, 3-6s, 6-8s blocks. Put an SFX cue on each one. The audio lands with it.
State the lighting and the motion. Fabric, water, hair, crowd. Omni animates what you name.
Then stop describing and start talking. Generate the base clip, and make every change a short instruction on the next turn. That is the talk-to-edit loop above.
Pricing
No Google account. No API keys. Generate with Omni Flash in the Coinis app and spend from one token balance. Live per-generation cost shows before you hit generate.
For context, Google's own API rate for Omni Flash is $0.10 per second of output.
Live now
Copy any prompt on this page. Paste it in the app. Then change the result by talking to it, turn by turn.
Generate with Google OmniFree to start. 25 tokens a week. No card.
Every clip on this page was generated with Gemini Omni Flash and finished with the Coinis endcard. Prompts shown as run. Gemini Omni is a Google model. Capabilities verified 2026-07-14.