The all-in-one local AI studio for your desktop: chat, image and video generation and a coding agent in one free, open source app. Windows and Linux. No Docker, no terminal, no cloud required.
TypeScript agentai-chatai-studioall-in-onecoding-agentdesktop-appfluximage-generationlinuxllmlocal-ailocal-llmopen-sourceprivacyself-hostedstable-diffusionuncensoreduncensored-aivideo-generationwindows
import { useCreateStore } from '../../../stores/createStore'
import { useCreateExp } from './CreateContext'
import { intentToJob, quotaCostKind, type RenderKind, type RenderOp } from '../../../lib/render/cloud-jobs'
import { lengthCredits, lengthPrice, runCredits } from '../../../stores/cloudCatalogStore'
import { createRunModel, createStudioCost, isStudioModel, startImageCount, studioIntentOf } from '../../../lib/render/create-studio'
import { STUDIO_MODELS } from '../../../lib/render/studio-contract'
import { runImageCount } from '../../../lib/render/image-count'
import { pictureStepCredits } from '../../../lib/three/picture-first'
/** What the cloud run on screen costs. One reader for the meter chip and for
* the list of several source images, so both name the same number. */
export interface RunCredits {
kind: RenderKind
op: RenderOp
/** One result. */
unitCost: number
/** Results per run (1 to 4 for Image and Edit). */
imageCount: number
/** One run: unitCost times imageCount. */
cost: number
}
/** Null while there is no quota yet, or no run to price. */
export function useRunCredits(): RunCredits | null {
const { quota } = useCreateExp()
const intent = useCreateStore((s) => s.intent())
const cloudImageModel = useCreateStore((s) => s.cloudImageModel)
const cloudVideoModel = useCreateStore((s) => s.cloudVideoModel)
const threeDWay = useCreateStore((s) => s.threeDWay)
const cloudOpModel = useCreateStore((s) => s.cloudOpModel)
const characterTab = useCreateStore((s) => s.characterTab)
const selectedCharacter = useCreateStore((s) => s.selectedCharacter)
// Review A kleiner Punkt 1: this meter is cloud-only (Composer renders it
// only for backend 'cloud'), so it must price the SAME length the cloud
// Length control shows and useCloudCreate books, cloudFrames/cloudFps, not
// the local track's frames/fps.
const frames = useCreateStore((s) => s.cloudFrames)
const fps = useCreateStore((s) => s.cloudFps)
const musicDuration = useCreateStore((s) => s.musicDuration)
const targetResolution = useCreateStore((s) => s.targetResolution)
const cloudStudioOptions = useCreateStore((s) => s.cloudStudioOptions)
const cloudStudioCredits = useCreateStore((s) => s.cloudStudioCredits)
const prompt = useCreateStore((s) => s.prompt)
const cloudImageCount = useCreateStore((s) => s.cloudImageCount)
const references = useCreateStore((s) => s.references)
if (!quota) return null
// No character picked yet: there is no run to price (resolveCharacterModel
// would fall onto the family's first model and show a number for a run
// nobody asked for). the web app fixed the same false reading.
if (intent === 'character' && characterTab === 'use' && !selectedCharacter) return null
let { kind, op } = intentToJob(intent)
// Mirror Composer's creditsOk pick exactly: meter and gate must show the
// same number: the character use-surface prices the family's real
// generation endpoint (never a hardcoded single default), the role
// intents (lipsync/music/extend/motion) run whichever Studio or classic
// op-picker model is chosen, everything else prices the per-kind
// picker's model.
const characterUse = intent === 'character' && characterTab === 'use'
if (characterUse) {
kind = 'image'
op = 'generate'
}
// Dieselbe Aufloesung wie Composer, Speicher und Start (createRunModel): eine
// Wahl, die die Unterkategorie nicht fahren kann, rechnet hier als das Modell,
// das wirklich laeuft.
const model = createRunModel(
studioIntentOf(intent, threeDWay),
{ image: cloudImageModel, video: cloudVideoModel, op: cloudOpModel },
characterUse ? selectedCharacter?.family ?? '' : undefined,
)
// Ein Studio-Modell rechnet nach seinem eigenen Schema, auch aus Image, Edit,
// Video und Animate (seit 02.10.2026).
const studioPick = !characterUse && isStudioModel(model) ? model : undefined
if (studioPick) kind = STUDIO_MODELS[studioPick].kind
const seconds =
op === 'music'
? musicDuration
: kind === 'video' && (op === 'generate' || op === 'animate') && fps > 0
? frames / fps
: undefined
// A Studio model prices from its own provider schema, never runCredits()
// (P3's guard there returns the caller's fallback on purpose, Portplan
// Abschnitt 7, Risiko 1). The Composer fetches the live number once
// (useStudioPrice) and leaves it here; this chip never asks twice for the
// same figure.
const extraPhotos = intent === 'edit' || intent === 'animate' ? references.length : 0
// Lip sync, voice, motion: the provider bills the length of the input. The
// Composer measures it and stores the price, and the same figure goes along
// with the start as its ceiling. Until then a voice counts its characters
// and everything else the catalogue length.
const lengthRate = studioPick ? undefined : lengthPrice(model, op)
const unitCost = lengthRate
? cloudStudioCredits ?? lengthCredits(model, lengthRate.basis === 'chars' ? Array.from(prompt).length || undefined : undefined)
: studioPick
? (cloudStudioCredits ?? createStudioCost(studioPick, cloudStudioOptions, Array.from(prompt).length, undefined, startImageCount(studioPick, extraPhotos) ?? 1))
+ pictureStepCredits(intent, threeDWay, studioPick, prompt)
: runCredits(kind, op, model, seconds, quota.costs[quotaCostKind(kind)], targetResolution)
// Mehrere Bilder: jedes ist ein eigener Auftrag zum vollen Preis, die Summe
// ist, was der Lauf bindet. Der Zaehler "noch N Bilder" rechnet je Bild.
const imageCount = runImageCount(intent, cloudImageCount, characterUse)
const cost = unitCost * imageCount
return { kind, op, unitCost, imageCount, cost }
}