Versely
    Guides

    Grok Imagine is the generate family; Grok 4.6 is the chat brain

    Imagine Image, Imagine Video, Extend, TTS are catalog rows. x-ai/grok-4.6 is chat_model. Pin the brain and the renderer separately or the look will drift.

    Versely Team••4 min read

    Grok Imagine is a generate family. x-ai/grok-4.6 is a chat brain. Pin them separately. If you only say “use Grok,” the agent may plan with 4.6 and render with Imagine Video — or plan with 4.6 and render with Kling. You will not know which knob moved.

    xAI’s August 12 Grok 4.6 announcement is about long-running agents in Cursor and Grok Build. Versely already had Imagine as pixel rows. The new slug on OpenRouter is how that agent brain enters our picker as org/model. It does not replace the Imagine pages.

    The generate rows (pixels)

    Live /models slugs, from the published catalog:

    • Grok Imagine Image — /models/grok-imagine — text-to-image, 4K, 2 credits. Fast, loosely styled stills.
    • Grok Imagine Image Quality — /models/grok-imagine-image-quality — text-to-image, quality tier up to 2K, 5 credits.
    • Grok Imagine Video — /models/grok-imagine-video — 6 / 10 / 15 / 20 / 30 seconds, 720p, native audio, 7 credits.
    • Grok Imagine Extend — /models/grok-imagine-extend — continue an existing Grok Imagine video, 2–10s, 8 credits, audio false.
    • Grok TTS — /models/grok-tts — text-to-audio. A voice file, not a talking generate.

    Those are renderers. They take a prompt (and, for extend, a clip). They spend generation credits. They do not plan a three-tool campaign.

    The chat id (plan)

    x-ai/grok-4.6: 500K context, tools, JSON, text + image + PDF in, text out, $2 / $6 per million on OpenRouter. Send it as chat_model with chat_model_explicit: true. Details: Grok 4.6 can see the still.

    Grok 4.3 remains a built-in vision chat id. 4.6 is the new OpenRouter slug. Check GET /api/v1/agentic/chat-models rather than assuming the UI grew a row.

    Chat credits and generation credits are two meters. Picking 4.6 changes the first. Picking Imagine Video changes the second. A 10-second Imagine Video at 7 credits is still 7 credits if the planner was GLM. A long 4.6 plan with no generate is still a chat bill with no MP4.

    /pricing is the live tariff. Do not treat OpenRouter’s $2 / $6 chat list as Versely credits. Those are different invoices: OpenRouter tokens versus catalog jobs. The agent’s done event returns credits_used for the reasoner and generation_credits_used for the renders. Read both.

    Four honest combinations

    1. Plan with 4.6, render with Imagine Video. You want xAI on both knobs. Say so. “Chat model Grok 4.6. Generate with Grok Imagine Video, 10s, 9:16.”
    2. Plan with 4.6, render with Seedance. You want the brain to see the pack shot, then hold the SKU in motion. Name Seedance 2.5 and /tools/ai-image-to-video. Do not let 4.6 “helpfully” route to Imagine Video.
    3. Plan with GLM 5.3, render with Imagine Image. Cheap text planner, xAI stills. No photo attachment (GLM is text-only). GLM 5.3 is a planner.
    4. Default Kimi, render with Imagine Quality. You did not pick a brain. You did pick a 5-credit 2K still. That is a valid week.

    The failure is the unstated mix: “use Grok” with a locked bottle still, and Imagine Video invents a second bottle because nobody named I2V.

    Extend is not 4.6 “continuing the thought”

    Grok Imagine Extend continues an existing Grok Imagine video at a timestamp, 6 or 10 seconds in the catalog copy, audio off. It is a generate tool. Grok 4.6 “continuing” is the chat loop: another tool call, another plan. Do not send a chat model into an extend endpoint. Do not ask Extend to write the next scene of a campaign.

    TTS is a voice file. A talking presenter is a lipsync or avatar row. Mixing those is how you buy a soundtrack you did not ask for, or a face you did not cast.

    FAQ

    Does pinning Grok 4.6 automatically use Imagine Video?

    No. The generate tool still needs a models argument. If you do not name one, a router picks. Name the row.

    Can I use Grok 4.6 only in Cursor and still publish on Versely?

    Cursor can hold 4.6 as its coding model. Versely jobs still need skills/MCP/CLI (npx skills add AI-XLabs-Innovation/versely-skills, or npx @versely/cli setup). The in-app agent pin is a separate chat_model.

    Is Imagine Video image-to-video?

    The published page is a video generator at 720p with native audio and those durations. For a locked still, prefer Seedance 2.5 + the image-to-video tool unless you have measured Imagine on that plate. Do not invent a folded /models/grok-imagine-image-to-video URL.

    Where is the roundup of other brains?

    9 chat models to pin behind the Versely agent.