Overview
Scribble uses generations, the same currency as the rest of Pencil. Every Scribble request draws from your subscription's generation allowance. There is no separate Scribble fee or add-on.
For how generations work across Pencil, read: How are generations counted?
How does Scribble use generations?
A Scribble turn is not a single model call. In one turn Scribble can plan the job, call several tools and models, and assemble the output. Each turn is priced on how much work it actually did.
When a turn finishes, Pencil places it in one of seven size buckets, from XS to XXXL, based on how many model calls it made and how much information those calls processed. The bucket takes whichever of the two is higher.
The rate for that bucket is set by the Chat model, which you pick from the Models menu in the composer. This is the model doing Scribble's thinking, so it prices the whole turn, the planning and reasoning as well as the reply. At the same bucket, a frontier model costs several times what a mid-range one does. For jobs that entail longer chat and brainstorming sessions with multiple back-and-forth turns, the Chat Model setting is the biggest lever you have over what each job and conversation costs overall.
Image, video and audio models are set separately in the same menu and are charged per asset, not per turn.
Chat is set to Auto by default, which lets Pencil pick the model that suits the job. Choose a model yourself when you want to hold every turn to a known rate.
The same menu carries a Mode setting. Adaptive, the default, lets Scribble judge how much effort each turn needs. Deep forces high effort on every turn, which produces more thorough work but costs more on average.
Two things follow from this. A quick copy tweak costs a fraction of a generation rather than a whole one. A full campaign brief is priced for what it actually involves, which is more per turn but far fewer turns end to end. You are not charged for failed generations.
How can I get the best value for my generations using Scribble?
Be mindful of how you prompt Scribble. Massive open-ended requests can result in runaway costs, especially if Scribble is given leeway to spawn multiple sub-agents and have them all work in parallel over long-horizon creative tasks. Here too the choice of Scribble's Default Chat Model can make a big difference.
Some tips:
Read the plan. Scribble shows you what it intends to do before anything runs. If the plan lists more shots, takes or variants than you need, say so before you approve it.
Say how much you want. Avoid scope creep and explicitly state the number of deliverables - concepts, variants, formats or videos of x seconds - that you require on each turn.
Ask for one first. On a brief you haven't run before, ask for a single variant, check it, then ask for the rest.
Check Chat Model settings. A frontier Chat model, and choosing Deep mode both raise the cost of every turn in the conversation. Using both at the same time is even more expensive.
As with other modalities, when generating creative with Scribble, remember that video is the costliest output format to generate, and is charged for each second generated.
For more information about the bucket thresholds, and the specific rates for every model, please consult How are generations counted?
What does a finished asset cost?
The cost of a finished asset is the agentic work plus any media Scribble generates along the way. The two are charged separately.
Adapting, resizing or cutting down an existing master creates no new media, so those briefs carry only the agentic cost. Generating new images or video adds the media rate for each image, or for each second of footage.
Your model choice moves the total a long way. A six-second clip costs 24 generations on Google Veo 3.1, at four generations per second, or 4.2 generations on Kling AI VIDEO 2.5 Turbo at 0.7. The creative decision and the cost decision are the same decision. On Auto, Scribble tries to balance cost and quality, picking the model it judges suitable for each step rather than defaulting to the best and most expensive one available. Scribble will generally reach for the leanest capable model, meaning the cheapest one that gives good results for that use case.
Scribble's capabilities and use-cases are expanding constantly. See this article for more about what Scribble can do today.
How do I track my generations?
The counter in the bottom right corner of your workspace shows your remaining generations. Admins can also view stats about overall generations used per user in any workspace via the Manage Team menu in Settings. Every Scribble turn and workflow step carries a cost menu showing the bucket it landed in, the generations charged, and the models used.
Start with the basics: What is Scribble and how do I use it?