MxChat Video Generation

AI video generation built into your WordPress dashboard, powered by Google Gemini Omni, ByteDance Seedance 2.0, and xAI Grok Imagine. Type a prompt, pick a model, get a finished MP4 — with native audio, multiple aspect ratios, and a Studio you talk to that turns one concept into a multi-scene film with consistent characters.

Overview

Most AI video tools live in a separate SaaS dashboard, behind a separate login, with their own media library. MxChat Video Generation puts the whole workflow inside wp-admin and saves every output straight to your WordPress media library — ready to embed, schedule, or hand to a client.

Two ways to work. The Generate tab is one prompt, one model, one clip. Studio is a workspace you talk to: describe what you want to make, and a Director drafts a treatment and scenes on a live storyboard canvas beside the chat, which you then render scene by scene or as a single multi-shot video.

Requires MxChat (mxchat-basic) with a Pro license active.

Setup

1. Install the plugin via WordPress admin (Plugins → Add New → Upload Plugin) or upload the ZIP.
2. Activate via Plugins → Installed Plugins. The plugin requires MxChat (mxchat-basic) and an active MxChat Pro license — you'll see admin notices if either is missing.
3. Get API keys for the providers you want to use:
- fal.ai API key (for Seedance 2.0 and Seedance 2.0 Fast) from your fal.ai account. Paste it into the fal.ai API Key field on the Settings tab and press Test key to confirm it works.
- Google Gemini API key (for Gemini Omni Flash) from Google AI Studio. Studio also uses this key for reference and scene image generation.
- xAI API key (optional — for Grok Imagine Video) from x.ai.
4. Navigate to MxChat → Video Generation in the WordPress admin sidebar (or Settings → Video Generation in network admin on multisite).
5. Open Settings, paste your keys, enable video generation, and test with a simple prompt like "A French bulldog running through a field of flowers."

Settings

Enable Video Generation

Master toggle. Off by default. Must be on before any of the Generate, History, or Studio screens will run a job.

Model Selection

Pick the active video model:

Gemini Omni Flash (Google) — clips of about 10 seconds with sound, 360p/720p/1080p, 16:9 or 9:16, from text or one reference image. Google's replacement for Veo, and the default on new installs.
Seedance 2.0 — native synchronized audio, multi-shot cuts, reference-image consistency. 4–15 seconds, 480p/720p/1080p, 16:9, 9:16, 1:1 or 4:3, up to 9 reference images.
Seedance 2.0 Fast — the cheaper, quicker tier. Use it for iteration and switch up for the final render.
Grok Imagine Video (xAI) — clips with audio, 480p or 720p.

Each option stays greyed out until its provider's key is saved.

Google retires Veo 3.1 on October 22, 2026, and its older Veo models are already retired. From 3.0.7 the Veo 2, Veo 3 and Veo 3.1 options are gone. A site that had a Veo model saved is moved to Gemini Omni Flash automatically and sees a one-time notice in WordPress admin; a Veo request from an old page or bookmark also runs on Gemini Omni Flash. See the troubleshooting entry below for what changes.

Aspect Ratio

Per-generation control, and the picker only shows ratios valid for the current model. Seedance 2.0 supports 16:9, 9:16, 1:1 and 4:3; Gemini Omni Flash supports 16:9 and 9:16.

Resolution

Seedance 2.0 runs at 480p, 720p or 1080p, defaulting to 720p — 480p is the cheap tier for test renders. Gemini Omni Flash runs at 360p, 720p or 1080p, defaulting to 720p.

Duration

Seedance 2.0: any length from 4 to 15 seconds, default 8. Gemini Omni Flash has no length setting: Google returns clips of about 10 seconds, so the duration is fixed.

Audio Toggle

On by default for models that support it (Seedance 2.0 and 2.0 Fast, Gemini Omni Flash).

Reference Images

Optional. Seedance 2.0 accepts up to nine reference images for character and style consistency; Gemini Omni Flash and most other models take one image.

Negative Prompts

Field for telling the model what to avoid (e.g. "no text overlays, no watermarks"), shown for Gemini Omni Flash and hidden for models that do not use it (3.0.8+). Omni has no separate negative-prompt setting, so the text is added to the end of your prompt as an "Avoid:" line.

Studio Models

Studio has its own model pickers, separate from the Generate tab. Video Generation Model offers Seedance 2.0 (the default) or xAI Grok Imagine Video. Image Generation Model — used for character references and scene images — offers Nano Banana 2.1, the default from 3.0.8 (Google's current fast image model, at about half the price of Nano Banana 2), Gemini 3.1 Flash Image (Nano Banana 2), the previous default, or Gemini 3 Pro Image (Nano Banana Pro) for higher quality. Sites that already picked a model keep their choice. A site that already chose Gemini 2.5 Flash Image keeps it, labelled as an older model Google no longer offers to new accounts, until March 15, 2027, Google's listed shutdown date; from that date Studio switches it to Nano Banana 2.1 and tells you once. From 3.0.7 a saved preview choice (Gemini 3.1 Flash Image Preview or Gemini 3 Pro Image Preview) moves to the current release of the same model, with a one-time notice. The Director writes scripts with your MxChat chat model; the Studio settings show which one under Script Generation.

Features

Studio is a workspace you talk to. Opening a project lands you in a chat panel beside a live storyboard canvas. Describe what you want to make; the Director drafts the treatment and scenes, updates them as you give notes, and everything appears on the canvas as you talk. There is no upfront questionnaire.
Nothing generates until you approve it. Anything that costs money — reference images, scene images, video renders — arrives in the chat as a confirmation card carrying a rough cost and time estimate. You approve the card, then it runs.
Every card is directly editable. Click the project title, logline, style, or any scene description to edit it in place. Regenerate or swap any reference image from its own card, and render or retry any single scene from its card. Aspect ratio, scene length and resolution are one-click chips on the treatment card.
Render the whole storyboard as one video. After approving your script and reference images, Generate as One Video renders the entire storyboard as a single multi-shot Seedance 2.0 video with synchronized sound, using your reference images to keep characters consistent across every shot — no per-scene clips and no stitching step. Scene-by-scene generation remains available as the advanced path.
Long runs survive the things that used to kill them. Scenes are submitted one at a time, so a hosting timeout no longer fails the whole run; a failed scene shows a Retry button and the exact provider error on its card; scenes that already finished are never regenerated or re-billed. Progress is saved to the project as it happens, so reloading the page or reopening the project resumes where it left off.
Export to Video Studio in one click. Once you have rendered scene clips, Export to Video Studio on the storyboard opens the hosted MxChat Video Studio timeline editor in a new tab with those clips already loading onto its timeline — trim, reorder, add music, and export one stitched film there, free. There is no download-then-upload step: the clips are passed as links and the editor fetches them itself. Only clips that finished rendering are sent, so a failed scene is skipped rather than blocking the hand-off. The Download videos button is still there for anything you would rather edit elsewhere.
Native audio generation: Seedance 2.0 and Gemini Omni Flash bake dialogue and sound effects into the clip during generation. No separate soundtrack step, no muxing.
Auto-save to media library: every generated video lands in WordPress media management as soon as generation completes. From 3.0.7 a finished Google video is saved to your site before it plays, so your API key is never part of the video link; the temporary copy is deleted a day later.
Works on a phone. Generation progress shows on the affected cards while the rest of the workspace stays usable, and on phones and tablets the chat collapses into a dock pinned to the bottom of the screen so the composer is always in reach.

API & Integration

The plugin talks to three video generation APIs and one image generation API:

fal.ai — Seedance 2.0 and Seedance 2.0 Fast, using your fal.ai key. This is the only route to Seedance; ByteDance's own API is not used.
Google Gemini API — Gemini Omni Flash (gemini-omni-1.1-flash), run in the background through Google's Interactions API so a long render does not hold the admin request open. The same Gemini key powers Studio's reference and scene image generation. From 3.0.7 the key is sent in the x-goog-api-key request header, not in the request address.
xAI Grok Imagine API — Grok Imagine Video, available as an alternative Studio scene-video model.

From 3.0.7, Seedance, Grok Imagine and Studio videos are downloaded only from your media library and the providers' own video addresses, over HTTPS. Seedance videos keep saving when fal.ai serves them from a new address on its video domain.

Studio projects persist in a custom table {prefix}mxchat_studio_projects with concept, config, script, reference data, scene images and video data columns, so a project survives a transient cache flush and can be resumed after a reload.

Requirements

MxChat (mxchat-basic): Required, active.
MxChat Pro License: Required — features are gated behind an active Pro license. See /pricing/.
WordPress: 5.8 or newer.
PHP: 7.4 or newer.
API keys: At minimum one provider key for the model you plan to use. Studio needs a Google Gemini key for images, plus fal.ai (Seedance) or xAI (Grok Imagine) for the scene videos.

Troubleshooting

My site was set to Veo and the model changed by itself

Google retires Veo 3.1 on October 22, 2026, and its earlier Veo models are already retired. From 3.0.7 a site with a Veo model saved is switched to Gemini Omni Flash, Google's replacement, and shown a one-time notice. It uses the same Gemini API key. Clips are about 10 seconds with sound, in 360p, 720p or 1080p, at 16:9 or 9:16, from text or one reference image. There is no duration choice, and options that only Veo had (Veo 3.1's three reference images, Veo 2's 5-second clips) are gone. If you need a different length or more reference images, switch to Seedance 2.0.

My site was set to Sora and the model changed by itself

OpenAI retired the Sora 2 API on 24 September 2026, so the Sora 2 and Sora 2 Pro options were removed in 3.0.5. Sites still set to a Sora model are automatically switched to Seedance 2.0 and shown a one-time notice. Add a fal.ai key on the Settings tab so the new model can run, or pick another model instead.

"API key not configured" or 401 / 403 errors

The provider key for the selected model is missing or invalid. Check MxChat → Video Generation → Settings and confirm the right key is filled in for the active model's provider. Google keys start with AIza.... For Seedance, use the Test key button beside the fal.ai field — it confirms the key in place rather than making you run a paid generation to find out.

Quota exceeded / rate limit errors

Every provider enforces per-account quotas and concurrency limits, and heavy Studio use hits them fast. Check your provider dashboard for current limits and billing status. Seedance 2.0 Fast and a 480p resolution are the cheap combination for iterating; move up for the final render.

A Studio scene failed

The scene's card shows a Retry button and the exact error the provider returned — read that first, it usually names the cause. Retrying re-renders only that scene; scenes that already succeeded are not regenerated and not billed again.

Generation hangs or times out

Generation times vary widely by model — roughly 30 seconds to 5 minutes for Seedance 2.0, and 15 seconds to 2 minutes for Gemini Omni Flash. Progress is saved to the project as it goes, so if a page times out, reopen the project and it resumes checking rather than losing the run. Completed videos also appear in the History tab even when the generation panel timed out.

There is no "stitch" button any more

There never really was one — the old interface promised browser-side stitching that never ran, and the message was removed in 3.0.5. There are now two honest ways to get one continuous piece. Generate as One Video renders the whole approved storyboard as a single multi-shot video. Export to Video Studio sends your already-rendered scene clips to the timeline editor, where you assemble them. Otherwise the download options are exactly what they say: individual clips, or a ZIP of every scene.

"Could not package your clips" when I press Export to Video Studio

The button collects the clips that have finished rendering, and there are none yet. Render at least one scene clip first, then export. Scenes that failed are skipped rather than blocking the hand-off, so you can export the ones that worked and retry the rest afterwards.

Export to Video Studio does nothing when I click it

It opens a new browser tab, so a popup blocker is the usual cause — the tab is opened on your click for exactly that reason, but some blockers are stricter. Allow popups for your admin domain and try again.

The Studio settings name Gemini 2.0 Flash as the script model

Fixed in 3.0.7. Gemini 2.0 Flash is shut down and was never what the Director used. The settings now show your MxChat chat model, which is what writes Studio scripts.

The progress ring, time estimate or character counter does not move

Update to 3.0.8. On the Generate tab these update again, and busy spinners show while a job runs.

The reference image browser will not pick, page or close, or a deleted video stays in History

Update to 3.0.8. The reference image browser works again for picking, paging and closing, and History deletes remove the video straight away.

Studio settings keep reverting

Fixed in 3.0.5. Saving the Studio image or video model used to rebuild the whole settings record and silently erase everything else in it, including your resolution choice — so a project could quietly render at the default resolution instead of the one you picked. Update to 3.0.5 and re-save your Studio settings once.

Last reviewed by Sage on 2026-10-25. Spotted an issue? Open a ticket and we'll patch the doc.