
The Engine Room
The right engine for every job. Free engines first. Our GPU is being switched on.
Engines differ: Pixar-style 3D, photoreal, text in images. One draws true vector logos. The OS lists about 100, live ones marked, and picks the right one per job, free first.
example dataWhat it is
No single AI is good at everything
One engine keeps a cartoon character consistent. Another makes real-looking photos. One writes clean text in images. Only one draws true vector logos (sharp at any size). 1 AI tool sold for everything is great at one thing, so-so elsewhere.
The OS lists about 100 named models, tools and engines, each marked live or planned. Each has 4 labels: license (client-safe?), speed and quality 1 to 5, cost, and lane.
Does it run while you're on the job? Yes. Paid engines run only by your choice. No mid-job price prompts: your spending limit works quietly, and over-limit jobs finish on a free engine.
The good-at map
Pick the look. The OS knows which engine makes it.
This is the short version of a long list. Tap any card to see the whole group.
The Seedream family holds a character's look shot to shot. Higgsfield Soul with a style lock. Adam was built this way.
Higgsfield Soul with an identity lock, Imagen 4 Ultra and Nano-Banana Pro. FLUX on our free engines for fast drafts.
Qwen-Image is the free engine first. Ideogram and GPT Image when every letter has to be exact.
Recraft is the only engine drawing real SVG: logos and brand marks sharp from a card to a truck wrap.
Wan 2.2 free. Kling, Veo and Higgsfield Cinema planned.
Talking-head engines are planned, with consent first.
Qwen3-TTS free (Kokoro planned). ElevenLabs on approval. Phone line: live once connected.
Claude on an owned writing seat goes first, $0 per job. Local models are the always-on backup.
The 7 groups
Every engine, sorted by the job it does
Free and owned engines go first in every group. The premium column runs only when you choose it.
| Group | How many | Free or owned, first | Premium, on approval |
|---|---|---|---|
| Text and writing | 9 | Claude (owned seat); Qwen3 8B and Llama 3.1 8B (local) | Claude API, a backup writer |
| Images | 25 | Qwen-Image, FLUX, SDXL, HiDream (free engines); Higgsfield Soul (owned seat) | Seedream, Ideogram, Recraft, GPT Image, Imagen 4 Ultra, Midjourney |
| Video | 24 | Revideo, FFmpeg, HyperFrames (local); Wan 2.2, LTX (free engines); Higgsfield (owned seat) | Kling, Veo, Sora, Seedance, Runway, Luma, Pika (planned) |
| Audio, voice and music | 14 | Qwen3-TTS, Kokoro, Chatterbox, ACE-Step | ElevenLabs v3; Suno (subscription) for songs with vocals |
| Avatars and talking heads | 6 | Adam SVG presenter (free); Higgsfield Soul ID and Speak (owned seat) | HeyGen Avatar III (planned) |
| Phone and voice AI | 7 | Phone line, live once connected | Vapi, Retell, Dograh, ElevenLabs Conversational |
| Search, SEO and data | 10 | Lighthouse, axe, pa11y, Firecrawl, SerpBear, Search Console | DataForSEO (tracks AI answers) |
Product fact: unique engines across the Engine Room manifest, model catalog, studios and back office. The video count includes the HeyGen and Higgsfield avatars, which show again under avatars. About 12 infrastructure tools (storage, workflows, email, booking, webchat, billing) are listed too.
The 4 lanes
Where each job runs, and who pays
| Lane | What it costs you | Examples | When it's used |
|---|---|---|---|
| Local | $0, always available | Ollama models, FFmpeg, loudness and noise cleanup (Kokoro and Piper planned) | The floor. If everything else is busy, this still works. |
| Free engines | $0 per job | Qwen-Image, FLUX, SDXL, Wan 2.2, LTX, Qwen3-TTS, ACE-Step | The default for images, video, voice and music. |
| Owned seat | Already paid | The Claude writing seat, Higgsfield | First pick where it beats the free floor, like photoreal with an identity lock. |
| Metered cloud | Per use, asks first | Kling, Veo, Sora, ElevenLabs, HeyGen, Seedream, Ideogram, Recraft, DataForSEO | Only when you choose it. Never the silent default. |
Product fact. When our GPU is off, hosted Replicate lanes can stand in; they're metered, so they ask first too. Suno, for music with vocals, runs on its own subscription. Metered engines need their key connected before they can run.
How a job is routed
Free first, every single time
Every job walks down the same ladder. Paid is never the silent default.
Primary
Each job type has a preferred engine; free is default. 8 job types: hero image, social image, 4K or print image, short video, long video, talking-head video, voiceover, music with lyrics and vocals.
Fallback
If the primary is busy or off, the next free or owned engine takes the job.
Paid, only if you chose it
A metered engine runs only when you picked it for this job, and it asks first.
Skipped
If no engine can do it within your limit, that part is skipped, logged, and its hold released. You aren't charged.
The router
The Model & Mode Router in Mission Control picks the cheapest engine that is good enough for the job.
The cost gate
Your spending limit, caught in the act
Every job has a cap and cheaper fallback. Set the limit once; Cost & Usage logs each time it kicks in. Here's how it looks, from the app's examples.
| What was about to happen | What the cap did |
|---|---|
| A video job projected at $5.40 | Capped at $5.00. The rest was made with Revideo, the free programmatic video engine. |
| A job projected to run past its GPU time | Split. The rest was upscaled 4x with Real-ESRGAN instead of rendered at full size. |
| A job projected at $5.10 | Stopped at the backstop. Free B-roll filled the tail. |
Sample: illustrations built into the Engine Room, not a client's bill. The billing rule (Product fact): quoted on the sum, charged on what returned, released on cached and skipped cells. Cost tracking shows real numbers once you go live.
The license gate. Each engine's badge says if its work is client-safe. Non-commercial-only engines are blocked before rendering, so nothing you post comes from a model you weren't allowed to use. Suno vocal music carries a rights flag.
Bake-off and Model Battle
Don't guess. Let them compete.
A bake-off runs 1 prompt on up to 6 engines; Model Battle pits 2 head to head. Plans, lanes and costs show first, free. Rendering waits for your OK.
Keep the winner. In the Engine Room you can change any job type's preferred engine. Free stays default; “Reset to free-first” restores all in one tap.
example dataThe proof
Adam: the same face in every feeling
Adam, our brand character, proves the right engine matters. A Pixar-style character must look like himself every frame, so his look was locked first and every render inherits it.
His library, as the app records it (Real): 72 chest-up expressions (12 families, 6 shades each, eyes-first) and 72 full-figure body-language poses. Same face, build, navy suit and red tie.
Your business can do this with your likeness, a team member, or a made-up host with AI disclosure, consent built in. See avatar engines and Avatar Studio.
12 families, 1 locked look
Real
Real
Real
Real
Real
Real
Real
Real
Real
Real
Real
RealWho does what
You, the OS and our team
- Set your spending limit once
- Approve any paid engine before it runs
- Keep the winner of a bake-off or battle
- Record consent for any face or voice
- Routes every job free-first, in the background
- Finishes on a free engine when a cap is hit
- Blocks non-commercial engines before any render
- Logs every route and every dollar in Cost & Usage
- Picks the best engine for every piece we make for you
- Runs bake-offs on big pieces and campaigns
- Keeps your look the same across engines
On Done-For-You, engine picks are our job. You see the finished work in your dashboard as we make it.
Go deeper
The 7 engine groups
Each group page lists every engine, its strengths, its lane, and where it works in the OS.
Questions owners ask
Do I need to know which engine to pick?
No. Every job type has a free default. Curious? The Engine Room's “Which engine for this?” button explains the pick in plain words.
Will I get a surprise AI bill?
No. Paid engines run only when you choose, and each paid use asks first. Your limit caps every job; overruns finish free. You're charged on what returned, not the quote.
What does “our own GPU” mean?
A dedicated graphics server rendering images, video, voice and music at $0 per job. It's the default visual lane, so most work runs free.
Do premium engines need their own account?
Some do. Metered engines need their key connected to run. The invideo editor runs on your own invideo plan and workspace, never a shared login.
Is the work safe to use in my marketing?
Every engine carries a license badge; non-commercial-only engines are blocked before any render. Anything with a person's face or voice needs their consent on file.
Why not use the most expensive engine every time?
The priciest isn't best at everything. Recraft alone draws true vectors; Qwen-Image is free and strong at in-image text. A bake-off compares them before you spend.
Keep going
Related pages
Try the OS
Our free trial is starting soon.
The same tools we use for our done-for-you clients. Our 7-day free trial is starting soon. Until then, book a call and a real person walks you through the OS. $997 a month, first month $497. No contracts.

YOUR TEAM runs it with the OS, or OUR TEAM runs it for you with Done-For-You. Not sure? Which one fits me?




