Auto, Claude, GPT, Gemini or Grok: the brain behind your AI director

Every production in Ask runs on two kinds of models. The brain is the one your AI director thinks with: it talks with you, asks what it needs, writes the storyboard and plans the renders. The engines do the rendering: Seedance, Kling, Veo and the rest of the lineup turn that plan into keyframes, clips and voice lines, one render per row. You pick the brain in the Ask bar; the engines are set in the plan, render by render.

This guide is for anyone who has opened that picker and wondered whether to touch it. The short answer: start on Auto, eroq's own pick for producing, which is included. Claude, GPT, Gemini and Grok are one click away when you want another voice in the writers' room, at a price per message you see before you send. New to Clap? Meet the AI producer first.

Two kinds of models, two different jobs

	The brain	The engines
Its job	Asks, writes the storyboard and every scene prompt, proposes the plan	Render the keyframes, clips and voice lines
Where you choose	The picker in the Ask bar	On the plan, render by render, or by naming one ("with Kling")
What it costs	Auto: included. A named model: credits per message	Each render's price, paid when you click "Approve · N credits"

One name sits on both sides, which makes the point: Grok 4.7 in the picker is a brain, Grok Imagine on a plan row is a video engine. Picking the first doesn't choose the second. The brain writes the plan; each row names the engine that renders it.

Here is one way a production on the defaults can go:

• You type: "a 15-second vertical ad for AMBRE, for TikTok".
• Clap asks one question, who's in it, and the chips are your own characters with their pictures, say Mara and Kai.
• You tap Mara. Clap writes a three-scene storyboard and proposes a look test. The plan card reads "Plan · 3 renders", from "Keyframe · Scene 1" to "Keyframe · Scene 3", 10 credits each.
• One button: "Approve · 30 credits".

The question, the storyboard and the scene prompts came from the brain. Every price on the card came from the studio.

The brain: what Clap thinks with

The picker sits in the Ask bar, on the studio home and in the chat, and applies to every message you send Clap. Open the picker and its first line sets the terms: "Auto is included. Pick another model to think with: it costs credits per message, shown before you send." Then the list, 6 brains in all, Auto included:

Brain	Per message
Auto	Included
Claude Sonnet 5.5	11 credits
Claude Opus 5.5	22 credits
GPT-5.6 Sol	11 credits
Gemini 3.1 Pro	13 credits
Grok 4.7	9 credits

Auto is eroq's own pick, and its row says so: "Our pick · included". It is built for producing in eroq, and it's where to start.

The named models each read "Billed per message · Auto steps in if it cannot answer", price on the right. Pick one and the pill in the bar shows its name and price per message, so the cost is in front of you before you press Enter. It stays picked until you change it.

Whichever brain you pick, it's still Clap at the table. It answers in your language, writes the engine prompts in English, follows the same rules and reads the same production: the brief, the cast, the elements, the storyboard and which renders landed. Like the rest of Ask, the picker lives in the signed-in studio only, not in the API and not in MCP.

The engines: what renders each shot

Engines aren't in the picker. Each row of the plan names its render ("Keyframe · Scene 1", "Clip · Scene 2", "Voiceover") with its model, length, format and price. Click a row and it opens on its full prompt, with a link to open that render in its own tool instead.

Unless you name an engine, or ask for something that needs another one, the plan uses the defaults: keyframes on the default image model at 10 credits each, clips on Motion One, eroq's own engine, at 145 credits for 5 seconds or 290 credits for 10, voice lines at 3 credits per 100 characters. Say "make the clips with Kling" and, if your plan unlocks it, the clip rows move to Kling 3.0 at its own price. How Ask reads engine names covers families and variants.

Now the rule that matters for your budget: the renders cost the same whatever brain planned them. The brain writes in names (Mara, vertical, Kling) and the studio turns them into the plan you see: it keeps only engines that are live and unlocked for your plan, fits every length to what the engine renders, and prices each row with the same function its tools charge with. The price on the button is the price charged, nothing renders before you approve, and a render that fails is refunded automatically. Approve before you render walks through the plan card line by line.

What a named brain costs, message by message

Say you're a few messages into a three-scene noir short with Mara, on Auto, and the keyframes have just landed: a grey line in the chat reads "3 renders landed". Before you spend on clips, you want a second opinion. You pick Claude Opus 5.5 and type: "What would you change in this storyboard? Don't rewrite it yet." The reply comes back like any other, with one more line under it: "This message used 22 credits." Then you switch back to Auto, tap a suggestion under "Next steps", say "Render all the clips", and your messages are included again.

If the model you picked can't answer a message, Auto answers it, the reply says so, and that message costs nothing. And a named brain changes what a message costs, never what a render costs.

For scale: ten messages on Claude Opus 5.5 come to 220 credits, ten on Grok 4.7 to 90 credits, and one 5-second clip on the default engine is 145 credits. Plans and top-ups are on the pricing page.

When a named brain is worth it

Start on Auto: it's built for producing in eroq, a clear request gets a storyboard and a plan straight away, and the renders cost what they would with any brain. A named model is a matter of taste and workflow, not a shortcut to better clips. Three cases where it can earn its credits:

• Another writing voice. Every brain gets the same instructions: director-grade scene prompts of 50 to 110 words, in English, with the same wardrobe and the same light words in every scene. Each model still phrases a scene its own way; if you want yours worded differently, try a named model on one message.
• A long film you keep reworking. A six-scene short can take many rounds of notes. If you already write with one of these models elsewhere and like how it takes them, keep it for the whole production, and watch the line under each reply: every message is billed. From storyboard to Cinema shows a full film session.
• A second opinion on a storyboard. That's the example above: one message, notes rather than a rewrite (a rewrite replaces the storyboard in the Production panel), then back to Auto.

FAQ

Which AI model is best for writing video prompts?

Start with Auto, eroq's own pick for producing, which is included. Every brain in the picker gets the same instructions for scene prompts (English, 50 to 110 words: who, where, shot, light, motion), so what differs is wording and taste. Try a named model on one scene and keep the writing you prefer.

Claude vs GPT for AI video: which should I pick?

They plan, they don't render, so your clips cost the same with either. Claude Sonnet 5.5 is 11 credits per message, Claude Opus 5.5 is 22 credits and GPT-5.6 Sol is 11 credits. Ask each for notes on the same storyboard and go with the voice you prefer.

Do I pay for Ask messages on Auto?

No. Auto is included: a message on Auto costs no credits. You spend credits on renders, when you click "Approve · N credits", and on messages to a named model, at the price the picker shows.

Do renders cost more when a named model plans them?

No. The studio prices every row of the plan with the same function its tools charge with, whatever brain wrote it. The total on the Approve button is what you're charged, and a failed render is refunded automatically.

What happens if the model I picked can't answer?

Auto answers that message instead, and it costs nothing. A small line under the reply says so: "Answered by Auto — the model you picked could not, so this message cost nothing." Your pick stays as it was for the next message.

Can I switch brains in the middle of a production?

Yes. The picker applies to the next message you send, and every brain reads the same production: brief, cast, elements, storyboard and the renders so far. Switch for one message, then back to Auto, as often as you like.

Open the studio, leave the picker in the Ask bar on Auto and tell Clap about your first production. Action!