The "AI model" tag on a Faymas prompt is a provenance label, not a compatibility guarantee. It records which assistant the creator says produced the sample image hanging next to the prompt text. It does not mean the prompt will run on that model inside the Faymas studio, because the studio names its models differently: Nano Banana, Veo, Wan, GPT Image 2, Kling, Seedream, Seedance, HappyHorse 1.1 and Grok Imagine. Not one of them is called "chatgpt" or "gemini". Treat the tag as a hint about where the sample came from, then choose the model yourself.
This guide analyses published prompt metadata and publicly listed studio models. No generation tests were run.
The tag answers "where", not "will it work"
Every prompt card on faymas.in carries a single model tag. We pulled the live homepage and a category page on 30 September 2026 and counted the tags on the twenty cards that render server-side. The result was lopsided: seventeen cards tagged chatgpt, three tagged gemini. Two values. No midjourney, no flux, no veo — even though Faymas's own marketing copy lists Midjourney, Veo, Runway, FLUX and Kling among the tools its prompts cover.
That gap is the whole story. The tag vocabulary is small and stable while the platform's stated model coverage is wide. So the tag is not a catalogue of everything Faymas supports. It is the creator's answer to one narrow question: which tool did you have open when you made this?
Which is useful. Just not in the way most people read it.
Two vocabularies that never meet
Faymas runs two surfaces, and they do not share a model vocabulary. The marketplace at faymas.in tags prompts with assistant names. The generation studio at faymas.net lists model engines. Here is what each surface actually says, as published on the day we checked.
| Surface | What it shows | Values observed |
|---|---|---|
| Marketplace card (faymas.in) | One Ai Model tag per prompt | chatgpt, gemini |
| Studio Model Matrix (faymas.net) | Nine selectable engines with one-line descriptions | Nano Banana, Veo, Wan, GPT Image 2, Kling, Seedream, Seedance, HappyHorse 1.1, Grok Imagine |
| Studio pricing page (faymas.net) | Seven engines named inside plan features | Veo3.1 & Veo3.1 Fast, Seedance, Kling V3.0, Nano Banana 2, Nano Banana Pro, GPT Image 2 |
Nine on one page. Seven on another. And neither list contains the two words the marketplace uses most.
So when you copy a prompt tagged chatgpt and walk into the studio expecting a button labelled ChatGPT, you will not find one. You will find GPT Image 2, which is the plausible OpenAI-side counterpart, and Nano Banana, which is the plausible Google-side counterpart. Plausible. We want to be precise here: that mapping is our reading, not something either page states. Neither Faymas surface publishes a translation table. Anyone who tells you the mapping is official is guessing.
Three things the tag does not tell you
It does not predict portability. A tag of gemini says the sample came out of a Google assistant. It says nothing about whether the prompt text survives contact with a different engine. Prompt syntax is model-specific in exactly the places that matter most — reference-image handling, aspect ratio flags, negative prompting. If you need to move a prompt across tools, that is a rewriting job, not a tag-reading job. We covered the rewriting side separately in our guide to adapting a Faymas prompt for other tools.
It does not predict cost. This one is easy to miss. The studio sells credits, and credit burn depends on which engine you pick and whether you are making a still or a clip. From the published plans you can work out the rough exchange rate: Starter is 220 credits a month and is described as up to 220 images or up to 22 videos, Pro is 500 credits for up to 500 images or 50 videos, Max is 2,000 credits for up to 2,000 images or 200 videos. Divide, and you get about one credit per image and ten credits per video. The tag on the marketplace card plays no part in that arithmetic. A prompt tagged chatgpt costs whatever the engine you actually select costs.
It does not certify the sample. The tag is creator-supplied metadata. There is no visible verification step that re-runs the prompt and confirms the label. The paired output image is the real evidence. The tag is a caption on that evidence.
Why the two studio lists disagree
The Model Matrix and the pricing page name overlapping but non-identical sets. The matrix lists Wan, Seedream, HappyHorse 1.1 and Grok Imagine; the pricing page does not. The pricing page lists Veo3.1 Fast, Nano Banana 2, Nano Banana Pro and Kling V3.0 with version numbers; the matrix uses bare Veo, Nano Banana and Kling.
The most likely explanation is boring: the two pages were written at different times and one of them is stale, and versioned names are the newer convention. We cannot confirm which. What we can say is that a platform which publishes two different model rosters on two pages you are both expected to trust during signup is a platform where the model list is not a stable contract. Plan around that. Do not build a workflow that depends on a specific engine still being there next quarter.
A thirty-second reading rule
You do not need to resolve any of this to use the site well. Apply this instead, in order.
- Read the output image first. Ignore the tag. Does the attached result look like what you want? That is the only claim being made.
- Read the prompt for tool-specific syntax. If it addresses an assistant in conversation — "use the uploaded photo as the main face reference" — it assumes an upload flow. Your engine needs to accept uploads.
- Pick the engine by output type, not by tag. Still image or clip? Photoreal or stylised? Need reference-image control? The studio's one-line descriptions answer those questions. The tag does not.
- Check the credit estimate before you generate. The studio shows an estimate before creation. That number, not the tag, is your real cost signal.
- Assume the roster moves. If a workflow matters to you, keep your prompt text in your own notes. The model underneath it is replaceable.
What we checked, and what we could not
Verified by direct fetch on 30 September 2026: the twenty marketplace card tags and their split (17 chatgpt, 3 gemini); the nine-engine Model Matrix on the studio homepage; the seven engine names and three plan tiers on the studio pricing page; the placement of the model tag inside a hover-only overlay in the marketplace markup, which means on a touch device with no hover state the tag may never surface at all — we did not test this on a real phone, so treat it as an observation about the markup rather than a confirmed mobile behaviour.
Could not verify, and so not asserted anywhere above: a per-model credit table (none published on either page); any official mapping between marketplace tags and studio engines; whether the tag is editable after publication or locked at submit time; and the identity of Nano Banana and GPT Image 2 as specific vendor models, which we have deliberately left as inference.
The short version
The tag is a receipt, not a promise. It tells you where a stranger got a good result. It does not tell you that you can get the same result, on the same engine, at the same price.
Which is fine. Receipts are still worth reading — they tell you whose kitchen the dish came out of. Just do not confuse the receipt for the recipe, and do not expect the kitchen to still be serving it in six months. Pick your engine, watch the credit estimate, and keep your own copy of anything that works.