How to run several AI models as an agency

We read the limits in our code. Unlimited personas on a paid plan, but one face per photo and no team seats.

Head of Content · Updated November 20, 2026 · How we test
The short answer

Keep every AI model as her own persona under one PixelPig account, generate for one persona at a time, and sort the output with the Gallery's persona filter. A paid plan has no persona cap. Pro connects 5 social accounts and Scale 15. Keep one model per account set, and get written consent for any real face you turn into a client's persona.

Create your AI model

To run several AI models as an agency, give every model her own persona under one PixelPig account, produce in batches one persona at a time, and keep a written map of which social accounts belong to which model. A paid plan has no persona cap, so the roster limit is your team's attention, not the software. Each model still needs the groundwork in how to create an AI influencer, and her accounts need the posting habits in growing an AI model on Instagram from zero.

Proof

We did not survey agencies for this page. The proof is the roster limits as the PixelPig source enforces them today, read from the persona, generation, Gallery, training, social and legal code. Where a limit could surprise an agency, it is in this table.

What an agency needsWhat PixelPig does today
Many models in one placeNo cap on personas for an account with an active plan. Without a plan you get one persona.
Several models in one photoNot supported. Every image and video request takes one persona.
Finding one model's outputThe Gallery filters by persona and shows a count per persona. Also filters by type, searches prompt text, and has a Favorites view. No folders.
Trained models (LoRA)One LoRA per persona, from 30 to 40 photos, on Pro and Scale. Default limit of 5 trained LoRAs per account.
Posting from one placeThe Content calendar connects Instagram, TikTok, X, Facebook and Threads. Pro: 5 accounts. Scale: 15.
Linking an account to a modelNot stored. A connected account has no persona field.
Working from ChatGPT or ClaudeThe connector lists your personas and calls one by exact name, up to 10 photos per call, on Pro and Scale.
Daily volume through the connector500 images and 100 edits per connector key per day, and 50 videos per account per rolling 24 hours, a count that includes videos made in Create. Every persona on the account shares these.
Photos per run in CreateUp to 4 per generation.
Ready-made facesNot exclusive. Taking one hides her from your own picker only; other accounts can still pick the same face.
Delivering files to clientsA file that carries a signed Content Credentials (C2PA) record downloads byte for byte with it. Files without one download with their metadata stripped.
Staff loginsNo team seats. Terms section 2 asks you not to share credentials.

Four rows shape the whole workflow. One face per generation means you batch production by model. The missing account-to-persona link means the map lives in your own notes. The LoRA limit of 5 means you choose which models get a trained model first. And shared ready-made faces mean a client who needs a face nobody else can use should get a persona built from consenting photos, not one from the picker.

We re-checked every row against the source on 20 November 2026. Nothing shipped since the first version of this page adds team seats, workspaces, multi-persona shots or a persona field on social accounts.

Step 1: write a brief for each model before you build her

Without written briefs, models tend to drift toward the same face and the same captions. Before building a persona, write one page per model:

  • Face spec: the features, expression range and styling, in one line each.
  • Niche and format: what she posts, in two words, from the niche ideas guide.
  • Accounts: which platforms and which handles belong to her.
  • Client and consent: who the model is for, and, for a real face, where the signed consent lives.

The brief is also your defence when a client asks why a model looks or sounds different this month. You can point to the page and the date it changed.

Step 2: build one persona per model

In PixelPig every model is a persona, and there are two ways to start one: pick a ready-made AI model, or use your own photos of a consenting adult, with a front-facing face photo required and portrait and full-body photos optional. The AI persona page shows both routes.

Pick the route per client. A ready-made face is fast, but it is shared: the picker hides a model only from the account that took her, so a second agency can run the same face. For a client whose model is the brand, build from photos the client owns or has licensed, with consent on file. What a virtual influencer is explains why the face is the asset.

Name each persona exactly as your brief names the model, because the ChatGPT and Claude connector finds a persona by its exact name, ignoring case, and uses one match if several names fit. Three naming rules follow from the code:

  • Give every persona a unique name. Two personas called "Mia" means the connector picks one of them.
  • Use letters, digits and spaces only. The lookup treats _ and % as wildcards, so "Mia_1" can match "Mia 1" or "Mia-1".
  • Keep names to 40 characters, the persona name limit. You can rename a persona later if a client renames the model.

If you build a persona from a client's own face, the Terms (section 4.2) require consent from every person shown, including yourself, that covers processing by an AI system. The acceptable use policy (section 3) adds that consent must be current and provable, and that permission to become a persona is not permission for adult output. Keep the consent with the brief, not in someone's inbox.

Step 3: batch production by model

Because each generation takes one persona, the fastest rhythm is one model per sitting: pick her in Create, make her week of photos and clips, then switch. Switching persona between every image costs focus and leads to mislabelled files.

If your team works from a chat assistant, the steps in making AI influencer photos inside ChatGPT or Claude apply per model. Ask the assistant to list personas, then request a batch for one name at a time. The token cost shows before anything runs, and every result also lands in the Gallery.

On Scale, Autopilot turns one plain-language brief into a plan for one persona and makes up to 10 safe-for-work images after you approve it. It only offers personas that already have a trained LoRA, which is one more reason to decide early which 5 models get one. The guide to how many photos you need to train a model of a person covers the training set.

Every image and clip lands in one Gallery, so the persona filter is your sorting desk. A working routine:

  1. Filter by one persona.
  2. Favorite the posts that match her brief.
  3. Delete the rest, so the next person who opens the filter sees only approved work.
  4. Download or schedule from the favorites. Where a file carries a Content Credentials record, the download keeps it. Either way, the AI label in the brief still applies.

The Academy chapter on content at scale uses the same filter-and-cull loop for a single model. With a roster, the cull matters more, because a stray image of the wrong model scheduled to the wrong account is an easy mistake for a client to spot.

Step 5: map accounts to models

The Content calendar connects each social account once, through the platform's own sign-in screen. Pro holds 5 connected accounts and Scale holds 15. Since PixelPig does not store which persona an account belongs to, keep that map in the brief and name the accounts so the mapping is obvious at a glance.

The Academy chapter on scheduling and posting sets two rules that matter more with several models: one model per set of accounts, and no identical file posted to many accounts at once. Near-identical posts across accounts look like spam to the platforms.

Disclosure applies to every model on the roster. Read how the Instagram AI info label works and TikTok's AI content rules before a client's first post, and write the labelling rule into each brief.

Solo creator vs agency setup

One model, one creatorSeveral models for clients
PersonasOneOne per model, named to match the brief
ProductionWhenever there is an ideaOne model per sitting, from the brief
Consent recordsYour own face or a ready-made modelA signed consent per real face, kept with the brief
Trained LoRAsOneUp to 5 by default, chosen by priority
Social accountsUp to 5 on ProUp to 15 on Scale, with a written account map
Output rightsNot claimed by PixelPigTransferable to clients under Terms section 4.4

Common mistakes

  • Sharing one login across the team. There are no seats, and Terms section 2 asks you not to share credentials. Section 1.2 notes that agency customers may be covered by a separate written agreement.
  • Opening extra PixelPig accounts to get around a plan limit. The AUP (section 5) forbids multiple accounts used to evade usage limits or tier gates. Do not open accounts in clients' names either: Terms section 2 allows an account for someone else only for an organisation you are authorised to represent.
  • Reselling access to PixelPig itself. Terms section 11.2 and AUP section 5 forbid pass-through access to the Service without independent value added. Selling the finished work is different: section 4.4 lets you pass outputs to clients.
  • Passing AI output off as human-made. Terms sections 5.5 and 11.4 forbid it, and AUP section 7 puts any legally required AI disclosure on you. The platforms ask for AI labels too.
  • Training a LoRA for every model on day one. With 5 by default, train the models that earn it first.

Bottom line

PixelPig handles the roster side well: personas are uncapped on a paid plan, the Gallery sorts by model, and Scale connects up to 15 social accounts. What it does not do is multi-face shots, team seats, or a stored link between accounts and models. Cover those with written briefs, one model per sitting and a clear account map. Compare plans on the pricing page.

Questions

How many AI models can I run from one PixelPig account?
There is no cap on personas for an account with an active plan. The limit that does exist is on trained LoRA models, which default to 5 per account on Pro and Scale.
Can my team share one PixelPig login?
No. PixelPig has no team seats, and section 2 of the Terms says not to share your credentials. Section 1.2 notes that agency customers may be covered by a separate written agreement.
Can I put two of my AI models in the same photo?
Not in one generation. Every image and video request takes a single persona.
Do I own the photos I make for a client?
PixelPig claims no ownership of your outputs, and Terms section 4.4 lets you transfer or sublicense them to clients. Whether AI output carries copyright depends on the country. Content you create for a client is not used in PixelPig marketing (section 4.6).
Can I turn a client's own face into a persona?
Yes, if that person gave consent that covers processing by an AI system and you can show it. Under the AUP, consent to become a persona is not consent to adult output.
How many social accounts can I connect?
Pro connects up to 5 social accounts and Scale up to 15, across Instagram, TikTok, X, Facebook and Threads. Connected accounts are not tied to a persona, so you keep that mapping yourself.
Which plan suits an agency?
Scale. It adds the API, Autopilot and the higher social account cap on top of everything in Pro. Plan details are on the pricing page.

Create your AI model

Create your AI model

Start here

Read next