Questions, answered

AI Studio is a creative production platform for AI-powered visual work. You can generate and edit images, produce short and long-form video (microdramas, ads, film scenes), build cinematic shots with consistent characters and locations, run multi-step creative briefs through the Agentic Canvas, and assemble multi-layer compositions in Photobash Studio. You can work standalone for one-off pieces or organise everything inside structured projects with episodes, scenes, and shots.

From the left sidebar, expand Projects and pick a template — Microdramas for short-form serialised content, Movies for long-form. Click New project, name it, and the template seeds an initial node tree (project → episode → scene → shot). Open any node and run a tool: Edit Sessions for stills, Agentic Canvas for multi-step work, Model Garden to pick a specific model. Everything you generate inside the project is attached to the node where you ran it.

No. The Agentic Canvas accepts plain-English briefs ("A 30-second ad for a luxury sneaker, dramatic lighting, end on the logo") and a planner builds the prompt chain for you. Edit Sessions also has prompt suggestions and reference-image inputs, so you can point at examples instead of writing prose. As you work, the system pulls signal from the references, Contexts, and approved generations you keep — your style compounds the more you use it.

Yes. Every account starts with a personal organisation that's yours alone — every tool is available, you can build full projects, and ship work entirely solo. Add a team later by creating or joining another organisation and switching between them via the picker at the top of the sidebar. Your personal org keeps its own credits, projects, and settings.

Every generation consumes credits. You have two pools: subscription credits (refresh each billing cycle, do not roll over) and top-up credits (one-time purchases that never expire). The system spends subscription credits first, then top-ups. Cost varies by model and output — a still image is cheap, a 10-second high-fidelity video is expensive. Each model lists its credit cost in the Model Garden before you run it, and failed generations are refunded automatically.

A growing catalogue spanning image, video, audio, and editing — Imagen-class image models, Veo and Kling-class video models, FLUX family, Wan, Nano-Banana editing, ElevenLabs voice, and providers wired through Replicate, FAL, BytePlus, and Google Vertex. Open the Model Garden in the sidebar to browse the live list with descriptions, sample inputs, and credit costs. New models are added regularly.

Uploads: PNG, JPG, WebP, HEIC for images · MP4, MOV, WebM for video · WAV, MP3, AAC for audio. Outputs: PNG for images and MP4 (H.264) for video by default; downloads include the original-quality master files. There is no platform watermark on paid plans.

Yes. Drop references into Edit Sessions, Photobash Studio, or any Model Garden model that accepts them. References can be character refs, style refs, structural refs (poses, layouts, depth), or video for image-to-video and motion-transfer flows. Photobash Studio is the place to assemble multi-layer references — stack, mask, scale — then push the final composite into a generator as a structural anchor.

Most AI video tools give you a single prompt box and one model. AI Studio is built around production: a hierarchical project tree (episodes → scenes → shots), reusable Worlds and Contexts that keep characters and brand guidelines consistent across hundreds of generations, an Agentic Canvas that orchestrates many models in one brief, multiple generation tools side-by-side, and team workflows with approvals and per-tool access. The platform is model-agnostic — you pick the best model per shot, not per app.

A planner-driven workspace where you describe the outcome in natural language and a multi-step agent orchestrates research, generation, and evaluation across the platform's tools and models. You see the plan, can edit any step, and iterate. Use it for ads, treatment-to-storyboard, shot lists, or anywhere you'd otherwise stitch many tools together by hand. Outputs land back in your project at the node you launched from.

Build a shot list inside a scene node — each shot becomes its own child node with its own session, references, and approvals. Shots inherit the scene's Context (brand, look, characters), so prompts can stay short while continuity stays tight. You can run a single shot, batch several at once via the Agentic Canvas, or use the planner to expand a one-line scene description into a full shot breakdown.

Yes — that's what the Worlds and Contexts system is built for. Define a character once (look, wardrobe, personality, reference shots) inside the project's Worlds, and every generation that pulls that World gets identity-locked references automatically. For tighter likeness, stack a Photobash Studio composite (character ref + pose ref + scene plate) and push it as a structural input to the chosen model.

Continuity is held by three layers: (1) Worlds — characters, locations, vehicles, wardrobe, and props registered at the project level; (2) Contexts — brand guidelines, look-and-feel notes, and style references that attach to scenes and shots; (3) Approved references — the generations you mark as canon become anchors for downstream shots. Together they keep faces, sets, lighting, and grade aligned across hundreds of generations.

Yes. The Microdramas and Movies project templates are structured for long-form work: episodes contain scenes, scenes contain shots. Generate shot-by-shot, approve as you go, and assemble in your editor of choice (downloads are clean MP4 masters). Worlds and Contexts carry across the whole hierarchy, so episode 12 stays consistent with episode 1 without re-prompting.

Brief the Agentic Canvas with the script or treatment. The planner produces a shot list, generates frames per shot, lets you approve the storyboard, then runs each approved frame through an image-to-video model with motion guidance. You can intervene at any step — rewrite a shot, swap the video model, regenerate a frame — without losing the rest of the plan.

Yes — ad workflows are a first-class use case. Drop your brand guidelines into a Context (colours, typography samples, do/don't list), drop product photography into a Worlds entry, then brief the Agentic Canvas: "15-second vertical ad, hook in the first 2 seconds, end on logo lockup". The planner produces options across compositions and pacings, generates them, and you pick winners. Final renders come out at the resolution and aspect ratio you specify.

Yes. Style is driven by the model you pick (the Model Garden has photoreal, anime, illustrated, and stylised specialists) and by the references you feed it. To lock a look, save 3–10 reference frames into a Context and attach it to your scene — every generation in that scene will pull the look automatically. Mixing styles across a project is fine; the system doesn't impose a global aesthetic.

Camera moves come from two places: (1) the video model's native motion controls (pan, dolly, push-in, tilt — exposed in the model's input panel) and (2) motion-transfer flows where you upload a reference clip and the model copies its camera path. For locked-off composition, use a still keyframe + img2vid with a low-motion setting. For dramatic moves, the Agentic Canvas can plan camera language across an entire scene.

A Project is the top-level container for a piece of work — a film, an ad campaign, a microdrama series. Internally it's a tree of nodes (project → episode → scene → shot, with depth varying by template). Each node holds its own sessions, generations, references, chats, and tasks. Open a node and you see everything attached to it; open the project root and you see everything across the tree.

Worlds is the project's character-and-location bible. Inside a project, World Building lives at the top of the tree and holds element groups: Characters, Locations, Vehicles, Wardrobe, Props, Music, and any custom types you add. Each entry stores reference images, descriptions, and metadata that downstream shots can pull from to stay consistent. Build the World once, use it across every scene.

Contexts are reusable knowledge packs — briefs, brand guidelines, methodology notes, look-and-feel decks, character bibles. Build a Context once and attach it to any chat, generation, or project. The attached Context rides along as instruction every time the AI works for you, so you don't re-paste the brand guidelines into every prompt. Contexts live at the org level so the whole team can reuse them.

Add teammates to your organisation (Settings → Members → Invite) and they instantly see every project the org owns. Inside a project, multiple people can work in parallel — different members can run sessions on different shots at the same time. Per-tool access flags let you restrict who can run which tools (e.g. contractor on stills only). Every generation is attributed to whoever ran it.

Yes. There is no per-project lock — anyone with org access can open any node and run sessions concurrently. To avoid two people editing the same session simultaneously, the session list shows who has it open. For larger teams, split the work by node (each member owns a scene) and use the My Tasks list to track assignments.

Tasks can be created on any node and assigned to a member — they show up in that user's My Tasks list. Generations can be marked as Approved (the canonical version), Pending Review, or Rejected. Approved generations are what downstream shots reference for continuity, and what the Agentic Canvas planner uses as anchors. The status trail is visible to the whole team.

Yes. Every generation is preserved with its prompt, model, references, parameters, and the user who ran it — even after deletion you can view the audit. Variants of the same shot are grouped together so you can compare side-by-side, promote a different one to Approved, or fork from any prior generation as a new starting point.

Use the project tree as the spine: episodes contain scenes, scenes contain shots. Each shot node holds its own approved render, alt-takes, references, and any uploaded plates. Worlds elements (characters, locations, vehicles) are linked into shots so the same character in shot 12 and shot 47 actually pulls from the same source. Reorder or rename nodes anytime.

Quick guide. Stills with iteration → Edit Sessions. Multi-step creative briefs → Agentic Canvas. Pick a specific model and run it → Model Garden. Stack and composite references before generating → Photobash Studio. Conversation, brainstorming, text → AI Chat. Brand kits, character bibles, briefs → Contexts. You can mix tools inside the same project — outputs from any tool can feed any other.

Agentic Canvas is plan-driven — you write a brief and an agent orchestrates many tools and models to produce a result. Playground is the freeform sandbox for exploratory generation on an infinite canvas (good for moodboards, ideation). Edit Sessions is iteration-driven — a persistent workspace where references and history stay loaded so you can refine a single image over many turns.

Use Model Garden when you know exactly which model you want and you want full control over its inputs — no planner in the way. It's also the best place to compare models on the same prompt (cost, latency, output quality shown alongside each result) and to discover newly added models. For everyday creative work, the other tools select models for you.

A multi-layer compositing canvas for building rich structural references. Stack images, mask, remove backgrounds, scale and arrange layers — then export the composite as a single reference and feed it to a generator. Perfect for combining a character ref + pose ref + scene plate, or for assembling product placement shots before generation.

Yes — outputs from any tool live in My Generations and can be reused as inputs anywhere. Generate a character in Edit Sessions, drop it into Photobash Studio for a composite, push the composite into Model Garden's video model, then reference the resulting clip from the Agentic Canvas for the next shot. Nothing is locked to its originating tool.

Within the product, the Agentic Canvas's planner is the general-purpose workflow builder — you describe the outcome and it composes the steps. For org-level workflows (custom node templates per project type, custom Context structures, fixed shot-list scaffolds), Owners can define them in Organization Settings. Bespoke production pipelines for studios are an Enterprise engagement — contact sales.

Most common reasons: (1) insufficient credits — top up or upgrade; (2) the underlying model's content policy blocked the prompt or reference (the failure reason is shown next to the failed item); (3) a transient provider error — re-run usually clears it. Failed generations refund credits automatically. If you see a failure that wasn't refunded, ping support@galleri5.com with the inference ID.

Common causes: (1) weak character reference — use a frontal, well-lit headshot at high resolution; (2) high motion strength on img2vid — drop motion intensity or use a model tuned for character work; (3) prompt fighting the reference — keep textual prompts short when you have strong refs, let the references do the work. For critical hero shots, lock the face with a Photobash composite that uses the character ref as a structural input.

Five things, in order of impact: (1) better references beat better prompts — feed the model 3–5 high-quality refs; (2) pick the right model for the job (Model Garden lists strengths per model); (3) use Contexts so style and brand are loaded automatically; (4) generate at the target resolution rather than upscaling later when you can; (5) iterate inside a single Edit Session so the workspace builds on itself instead of starting fresh each time.

For characters: clean, well-lit headshots and three-quarter angles, neutral background, ideally 1024px+. For style: 3–10 frames that share the look you want — grade, lighting, lens. For composition: a rough blockout or a Photobash plate showing layout. Avoid noisy backgrounds, JPEG artefacts, and tiny thumbnails — the model will reproduce whatever flaws are in the ref.

Flicker is mostly a video-model problem. Workarounds: (1) use a model with strong temporal coherence (the Model Garden flags this per model); (2) lower motion intensity if your scene is mostly static; (3) generate shorter clips and edit them together rather than one long take; (4) lock the first and last frame with strong stills, then use frame-to-frame interpolation.

Three factors: model (top-tier video models can take several minutes per clip), provider queue (capacity spikes at peak hours), and input complexity (more references, longer videos, higher resolution all add time). Generations run async — close the tab if you want, they'll appear in My Generations when done and you'll be notified.

Run a generation at the model's native resolution, then use an upscaler from Model Garden (image or video upscalers are listed there with credit costs). Upscaling is non-destructive — the master is preserved. For delivery, generate a 1024 or 1080 master, upscale to 4K only at final stage; this saves credits during iteration.

Image: most models hit 1024×1024 to 2048×2048 natively, with non-square aspect ratios (16:9, 9:16, 4:5) supported per model. Video: 720p and 1080p are common; some models now hit 4K. Aspect ratio (vertical for ads, cinematic scope for film) is selectable per generation. Each model in Model Garden lists its supported resolutions and ratios.

Yes — production teams use AI Studio to run microdrama pipelines, ad volume work, and previs/concept work for feature productions. The combination of project trees, Worlds, Contexts, approvals, and per-tool access flags scales to studio use. For high-volume production, Enterprise plans add dedicated capacity, custom node templates, audit logs, SSO, and an SLA. Reach out via the Pricing page footer to talk through your pipeline.

Yes, on Enterprise plans. Private-tenant deployments give you isolated infrastructure, your own data residency region, and dedicated GPU capacity. We can also deploy into a customer's cloud account for stricter compliance needs. Contact sales for scope and timelines.

On Enterprise plans, yes — we onboard customer-owned checkpoints and LoRAs into a private slot in the Model Garden, accessible only to your organisation. We also support fine-tuning runs on customer datasets where the underlying provider permits it. Discuss model rights and licensing with sales before onboarding.

Programmatic access is available on Enterprise plans for automation use cases — running batches, integrating into studio pipelines, exporting deliverables. The web app's internal API is not currently a published public surface, so API contracts are scoped per engagement. Talk to sales about your integration shape.

AI Studio is built on Azure today (App Service + Container Registry) and Enterprise customers can deploy into their own Azure tenant. On-prem GPU clusters are possible for the inference layer; the application tier still benefits from being cloud-resident for updates and operations. Scope these via an Enterprise discussion.

Yes — Galleri5 runs producer-led teams that embed with studios for the run of a project: shot supervisors, prompt engineers, character consistency leads, and a production manager. Useful when you want guaranteed throughput on a deadline. Contact sales for staffing and engagement structure.

You own everything you upload and everything you generate. Galleri5 does not claim rights over customer outputs. Underlying-model providers each have their own commercial-use terms (most allow commercial use; some require attribution or restrict certain content); the Model Garden surfaces each model's terms before you run. Enterprise agreements include explicit IP and indemnity clauses.