Preferences

Preferences is where you decide which AI models the chat uses when it makes pictures and clips for this site, how large and long they may be, and what each costs. It sits behind More, just above Settings.

Everything here belongs to the project, not to you. A teammate opening the same site sees the same choices, a change made by one person shows up in everyone else's open editor straight away with no reload, and anything they ask for is charged at the same rate. That is deliberate: generated images are committed into the site's code and the credits come out of the shared workspace balance, so two people on different settings would be billing the workspace differently for the same site.

Every row saves as soon as you change it, and a toast says what new pictures or clips will use and cost, such as New pictures will use Nano Banana 2.1 up to 2K, at 0.37 credits each. on Starter. If a save fails, Could not save your preference gives the reason and the row goes back to what was saved.

Paid media

Allow paid images and video sits at the top because it governs everything below it, and it is on for a new site. Turn it off and the chat can still search free stock photography, which costs nothing, but it cannot generate, edit or animate. Every setting below it stays editable while it is off, so you can pick the models, sizes and clip settings first and switch paid media on once they are right.

It only affects this site, and turning it back on restores the choices you had made. Anyone with the Editor role on the site can also change it, so it is a switch for this site's spending rather than a lock that holds against a teammate.

Images

Two rows decide what a picture costs: which model draws it, and how large it may be.

Image model is the model. Each row names the model, with its provider and what it is good at on the line beneath, and shows its price at your current size ceiling before you pick it. The picker groups the models by who makes them, under a heading for Google, OpenAI, ByteDance, Black Forest Labs and Ideogram, with Recommended above the groups:

The three Nano Banana models are Google's Gemini image models, listed under the names Google gives them. Nano Banana 2 Lite is Gemini 3.1 Flash Lite Image and Nano Banana Pro is Gemini 3 Pro Image.

Under the picker, a line gives the price per image at your ceiling, what an edit costs when that differs, and whether the model takes any shape or square, landscape or portrait only.

Recommended is the default and it is not the same as picking Nano Banana 2.1. Recommended follows whatever Modulify recommends, so if the recommended model changes later, a site left on Recommended moves with it. A site pinned to a named model stays there. If a pinned model is ever unavailable, it drops out of the picker, which then shows Recommended, and the chat uses the recommended model until the pinned one is back.

Largest image size is a ceiling of 1K, 2K or 4K, and the default is 2K. The chat may render a smaller picture when that suits, but never a larger one, and anything it asks for above the ceiling is brought back down to it. The prices next to each model change as you move the ceiling.

Prices depend on your plan, because a Pro credit is worth twice a Starter credit. The same picture costs the same money on either plan, so a Pro workspace simply spends half as many credits on it. Free and Enterprise workspaces pay the Starter prices. Change plan with the editor open and every price on this tab updates on the spot.

FLUX.2 Klein tops out at 2K, so a site set to 4K gets a 2K picture from it at the 2K price, which is why its 4K column repeats the 2K one. While it is picked, 4K in the size row is greyed out with Not offered by this model. FLUX 3 goes up to 4K, where it costs far more than at 2K. Nano Banana 2 Lite always renders at 1K, and Seedream 5 Lite and Ideogram 3 Turbo set their own size from the aspect ratio, like the two GPT Image models, so the size row greys out for them with This model sets its own size from the aspect ratio. and their price is the same at every size.

Those are credits per picture, and they are charged exactly rather than rounded up to a whole credit. A square picture from either GPT Image model costs a little more than the prices above, which are for the usual wide shape: 0.33 credits on Starter and 0.16 on Pro.

Editing a picture costs a little more than making one on some models, because they also charge for reading the picture being edited. The difference is biggest on the two GPT Image models, where a wide edit is 0.34 credits on Starter and 0.17 on Pro, and a square one 0.41 and 0.20. On Nano Banana 2.1, Nano Banana Pro and FLUX.2 Klein an edit adds 0.01 credits at most. Nano Banana 2 Lite, Seedream 5 Lite and FLUX 3 edit at the same price as they generate. Ideogram 3 Turbo makes new pictures but cannot edit one, so on a site set to it, edits are made with Nano Banana 2.1 at that model's price, and the price line under the picker says so.

A batch multiplies the price, so six Nano Banana Pro pictures at 4K is 8.76 credits on Starter. A picture that fails is not charged.

When words appear inside the picture

The two GPT Image models are the ones to reach for when the picture itself contains writing, on a poster, a badge, a packaging mockup or an app screenshot. They cost the same. GPT Image 2.5 Flare is the everyday one. GPT Image 2.5 Sunburst is the one to pick when an edit has to change exactly what you asked and leave the rest alone.

They carry one trade off no other model does. Both render only three shapes, square, landscape and portrait, while the other models follow the aspect ratio. So asking a GPT Image model for a very wide 21

Ideogram 3 Turbo is the one to pick for lettering and logos. It keeps the wording sharp, follows every shape except 21

Picking by what the picture is for

Pick Nano Banana Pro when you want the richest result and the shape has to be exact.

Pick Nano Banana 2 Lite for quick pictures that do not need to be large. It always renders at 1K and costs 0.21 credits on Starter whatever the size ceiling.

Pick Seedream 5 Lite for heroes and product shots at a low price. It costs 0.21 credits on Starter whatever the size ceiling, so it is the cheap way to fill a page with rich pictures.

Pick FLUX 3 for photoreal work, a scene that has to look like a photograph rather than an illustration. It is 0.14 credits on Starter at 1K and 0.30 at 2K. It goes up to 4K, but at 1.82 credits there its picture is the most expensive on this tab, so keep the ceiling at 2K unless you need the detail.

Pick FLUX.2 Klein for icons, placeholders and drafts. At 0.01 credits for a 1K picture on either plan it costs almost nothing, so it suits trying a layout before spending on the final pictures. It goes up to 2K.

Video

Three rows decide what a clip costs, and they combine rather than being one fixed bundle. A fourth decides whether visitors can hear it, and on some models it changes the price too.

Video model is the model, and the picker groups the ten by who makes them, with Recommended above the groups. Each row shows its price at your current length and resolution before you pick it:

Seedance 2.0 buys the strongest motion, not a longer or sharper clip. Gemini Omni Flash, Google's video model, gives realistic motion and makes sound in every clip. Seedance 1.5 Pro gives smooth motion at a low price, and Seedance 1.0 Pro Fast is cheaper still but silent. Kling 3.0 and Kling 2.6 are the newer Kling models and add sound when you want it. Wan 3 and MiniMax H3 make sound in every clip at a low price. For a looping background, Seedance 1.5 Pro or LTX 2.5 Fast is the sensible choice. A pinned video model that is unavailable falls back to Recommended the same way as an image model. Under the picker, a line gives the price of one clip at your settings, such as 0.62 credits per clip, 4 seconds at 720p for Recommended on Starter.

Longest clip is a ceiling, and the default is 4 seconds. The lengths on offer depend on the model: 4, 6, 8 or 10 seconds on Gemini Omni Flash, 4, 6, 8, 10, 12 or 15 seconds on Seedance 2.0 and Kling 3.0, 4, 6, 8, 10 or 12 seconds on Seedance 1.5 Pro and Seedance 1.0 Pro Fast, 4, 6, 8, 10, 12, 16 or 20 seconds on LTX 2.5 Fast, 4, 6, 8, 10, 12, 15 or 20 seconds on Wan 3, 5, 6, 8, 10, 12 or 15 seconds on MiniMax H3, and 5 or 10 seconds on Kling 2.6 and Kling 2.5 Turbo. The chat may make something shorter when that suits, but never longer. When the model you pick does not offer your ceiling, the clip takes the nearest length below it, or the shortest that model makes, so a site set to 4 seconds gets a 5 second clip from Kling 2.6, Kling 2.5 Turbo or MiniMax H3.

Clip resolution is 720p or 1080p, and the default is 720p. Kling 3.0, Kling 2.5 Turbo, Seedance 1.5 Pro and MiniMax H3 render 720p only, MiniMax H3 at 768 pixels tall, and Kling 2.6 renders 1080p only, so the row offers nothing else while one of them is picked. On every other model Full HD takes any length the model offers.

Those are credits per clip, and a length the table does not show is priced per second the same way, so a 12 second Seedance 1.5 Pro clip with sound off is 1.87 credits on Starter. Kling 2.5 Turbo, at 5 or 10 seconds, is 1.26 or 2.52 credits on Starter, and 0.63 or 1.26 on Pro. Kling 2.6, at 5 or 10 seconds in 1080p, is 2.10 or 4.20 credits on Starter with sound off and 4.20 or 8.40 with it on, and half that on Pro. MiniMax H3 starts at 5 seconds, which is 1.80 credits on Starter and 0.90 on Pro, and adds 0.36 and 0.18 for every second after. On Seedance 2.0, Full HD costs two and a half times as much per second as 720p, so a 15 second clip at 1080p is 40.50 credits on Starter, the most one clip can cost. Seedance 1.5 Pro, Kling 3.0 and Kling 2.6 are the models whose price depends on Allow sound: a clip made with sound costs twice as much on Seedance 1.5 Pro and Kling 2.6, and half as much again on Kling 3.0, which is why they have two prices. Gemini Omni Flash, Seedance 2.0, LTX 2.5 Fast, Wan 3 and MiniMax H3 include sound at the one price.

Allow sound decides whether visitors can hear a clip. On Gemini Omni Flash, Wan 3 and MiniMax H3 every clip is generated with sound and the price already includes it, so the setting costs nothing either way. On Seedance 2.0, Seedance 1.5 Pro, LTX 2.5 Fast, Kling 3.0 and Kling 2.6 the switch also decides whether the clip is made with sound at all, and on Seedance 1.5 Pro, Kling 3.0 and Kling 2.6 it raises the price when on. Seedance 1.0 Pro Fast and Kling 2.5 Turbo make silent clips, so the switch greys out for them with the note This model makes silent clips. It starts off on a site where nobody has changed it. With it off, the clip plays muted with no way to unmute, which suits a looping background. With it on, the chat adds controls so a visitor can turn the sound on. Either way a clip starts muted, because browsers refuse to autoplay sound. On Gemini Omni Flash, Wan 3 and MiniMax H3, turning it off hides the sound on the page but does not remove it from the file, so anyone who downloads the clip can still hear it.

There is no 4K clip, and no clip from a text description alone, so animating always needs a picture to start from.

Asking for something different in chat

You do not have to come here to change things for one request. Say it in the message instead:

Make a hero image for the pricing page, use GPT Image 2.5 Sunburst

The same works for clips: "animate this with Kling 3.0" applies to that one clip. It does not change anything saved here, and the chat tells you the new price before it spends it.

What you ask for in a one-off request still cannot go past the ceilings on this tab. A request for a longer clip, a larger picture or sound that this tab does not allow is brought back to what the tab allows rather than overspending. To go further, raise the ceiling here first, or ask the chat to raise it for you.

Changing preferences from chat or an AI client

Everything on this tab can also be changed without opening it. Ask the chat in plain words and it saves the setting for the whole site:

Use GPT Image 2.5 Flare for images from now on, and allow clips up to 8 seconds

The chat reads the current settings and their prices before it answers, and it only changes a saved preference when your own message asks for that change. A request for one picture in a different model is not a preference change, so it stays a one-off. Anything that makes later pictures or clips cost more, such as turning paid media on, picking a pricier model or raising a ceiling, comes with the new price in the reply, and if the change was not clearly what you asked for the chat checks with you first. If paid media is off and you ask for a generated picture, the chat asks before switching it on, and the new setting applies from your next message.

An AI client connected over MCP can do the same with get_ai_preferences and update_ai_preferences. See MCP tools.

Whichever way a preference changes, this tab updates in every open editor on the site straight away.

Where the credits come from

Image and video credits come out of the same workspace balance as everything else the chat does, and they show up in your usage history under their own entries. A picture or clip is charged when it starts and refunded if it fails, so the balance can never run out partway through one, and you are only ever charged for what you receive. Credits covers the balance, the monthly allowance and topping up.