AI Video Model Selection Guide for Creative Disciplines.

Why Model Choice Matters More Than Prompt Quality

Pick the wrong video model and no amount of prompt engineering will save you. You will burn credits, miss deadlines, and end up with footage that almost does what you need. The model is the decision that shapes everything downstream.

Most tutorials spend 90% of their time on prompt structure and almost none on model selection. That imbalance is backwards. The practical decision tree looks like this: fix your output format first, match a model to that format, then refine the prompt. Not the other way around.

The format question covers three things. What is the input mode: text-to-video, image-to-video, or keyframe-controlled with defined start and end states? What are the constraints: clip length, aspect ratio, required resolution, budget per shot? And what kind of motion control do you need: camera movement, subject behaviour, or loop integrity?

A product designer generating hero videos for a pitch deck and a motion designer building a looping brand ident are both making AI video. But they need fundamentally different things from the model. The product designer needs surface fidelity and controlled reveals. The motion designer needs loop integrity and frame-accurate start and end states. Sending both workflows through the same model at the same settings is not a neutral choice. It is usually the wrong one for at least one of them.

The same logic applies across all twelve disciplines Stensyl serves. Film and set designers need camera continuity across pre-vis shots. Automotive designers need clean paint and reflective surface behaviour. Content creators need speed and volume. Exhibition designers need controlled spatial transitions. Each discipline has a natural model fit. The rest of this article maps those fits directly.

The Current Video Model Lineup: What Each One Actually Does

Stensyl's video generation surfaces, including the Video surface and the Film studio, currently expose four Luma models. Understanding what each one is built for prevents the most common credit-wasting mistakes.

Luma Ray 3.2

Ray 3.2 is the keyframe-controlled flagship. You supply a start frame, an end frame, or both, and it generates a clip that moves between those two visual states. Output is 5 or 10 seconds. Loop mode is supported natively, which makes it the direct fit for any discipline that needs seamless repeating motion: brand idents, exhibition display loops, product turntables, animated hero banners.

Because you lock the first and last visual states, Ray 3.2 gives you the highest degree of compositional control of any model currently on the platform. That is the reason to use it. It is not the cheapest option, so reserve it for shots where that control actually matters.

Luma Ray 2 Flash

Ray 2 Flash trades some fidelity for speed and lower credit cost. It is the drafting model. Use it when you need to validate motion, composition, or timing before committing to a full-quality render. For content and social workflows where you need multiple short clips quickly, Flash's speed-to-credit ratio is the main reason to reach for it.

The practical split is straightforward: draft in Flash, finalise in Ray 3.2. Running two or three Flash iterations first costs less than a single misdirected Ray 3.2 render.

Luma Uni-1

Uni-1 is an image generation model, not a video model. It generates images using reasoning-grounded prompts with strong typography handling. Its role in video workflows is as a frame-generation step: use it to produce a high-quality hero still, then feed that still into Ray 3.2 as a keyframe. This is a standard multi-stage pipeline approach and it is more controllable than prompting video cold.

What is No Longer Available

Luma Ray 2 (base) was retired in June 2026 and is no longer accessible on Stensyl. If you encounter tutorials or comparison articles that reference Ray 2 as a current option, those guides are out of date. The current lineup is Ray 3.2, Ray 2 Flash, and Uni-1.

Matching Models to Disciplines: A Practical Breakdown

The twelve disciplines Stensyl serves don't all have the same video output requirements. This section maps each major use case to the model and workflow that fits it.

Film and Set Design: Pre-Vis Sequences

The production-proven workflow here is: storyboard, generate keyframes, animate via image-to-video, then edit. Film and set designers generating pre-vis sequences for director review should render individual storyboard frames in Scene Composer or Uni-1, then animate them in Ray 3.2 using those renders as start and end keyframes. This gives you controlled camera-to-camera continuity across shots rather than hoping a text prompt produces the right framing.

For complex sequences, use the Film studio to sequence model calls across shots rather than manually chaining individual Canvas video nodes. Film handles multi-scene orchestration natively.

Product Design: Hero Videos for Pitches

Surface fidelity is the priority: labels, material finishes, reflections, and edge definition all need to survive motion. The practical approach is to build the hero still in Uni-1 or Scene Composer first, locking the composition and lighting before any video generation begins. Then draft angle variations in Ray 2 Flash to validate motion and framing. Once the composition is confirmed, run the final render through Ray 3.2.

Generating the video directly from a text prompt risks surface smearing on fine details. Locking a hero still first protects the design intent through the animation step.

Motion Design: Looping Brand Idents

Loop integrity is the core requirement here, and Ray 3.2 supports looping natively. The workflow is to design the first and last frames of the loop deliberately, ideally as separate still renders so the transition is visually intentional rather than inferred. Draft the timing in Ray 2 Flash, then commit to Ray 3.2 for the loopable final output. Export via the Motion studio for production-quality output.

Attempting loops through pure text-to-video prompting is unreliable across all current models. Keyframe-controlled generation is the correct approach for this use case.

Content and Social: Multiple Short Clips at Volume

For creators producing TikTok, Reels, or Shorts content, Ray 2 Flash is the primary tool. The format is typically fixed early (vertical aspect, 5–15 seconds), the content needs multiple variations, and speed matters more than maximum fidelity. Flash's lower credit cost per render means you can generate more options per session without exhausting your monthly allowance.

The Marketing studio feeds into Editing for caption overlays and final export, which keeps the surface count low for this kind of short-form social pipeline.

Game Development: Cinematic Trailers and Pitch Pre-Vis

AI video in game development is primarily used for mood boards, trailer pre-vis, and pitch presentations rather than final in-game cutscenes. The recommended approach is to build a continuity reference in Boards, grouping character poses, environment frames, and prop stills into start and end scene pairings. Render those through Ray 3.2 via Canvas using the Assemble Film node for multi-shot batch orchestration. The Film studio handles the sequencing natively if you are working with a linear shot list.

Automotive and Exhibition Design: Controlled Reveals

Both disciplines need controlled spatial or surface transitions. An automotive designer presenting a paint-finish reveal and an exhibition designer showing a stand environment unfolding both need the camera path and surface behaviour to be predictable. Ray 3.2 with a strong first frame from Scene Composer is the fit: you control the starting state, define the end state, and let the model fill the transition. Draft the motion in Flash to check pacing, then finalise in Ray 3.2.

Discipline	Primary Need	Draft Model	Final Model	Key Surface
Film and set design	Shot continuity across pre-vis	Ray 2 Flash	Ray 3.2	Film studio
Product design	Surface fidelity on hero video	Ray 2 Flash	Ray 3.2	Canvas / Scene Composer
Motion design	Loop integrity for brand idents	Ray 2 Flash	Ray 3.2	Motion studio
Content and social	Speed and volume of clips	Ray 2 Flash	Ray 2 Flash	Marketing studio
Game development	Multi-shot trailer sequencing	Ray 2 Flash	Ray 3.2	Boards + Canvas
Automotive / exhibition	Controlled surface or spatial reveal	Ray 2 Flash	Ray 3.2	Scene Composer + Canvas

Using Boards and Film Studio to Set Up Better Video Outputs

Model selection is only half the decision. The surface you use to set up the generation shapes the quality of the output as much as the model itself.

Boards for Keyframe Preparation

Boards is a single fluid canvas that combines visual reference collection with frame grouping for first and last-frame video generation. Instead of manually uploading separate keyframe files each time you start a new shot, you collect your references and arrange them into start and end scene pairings directly on the canvas. Those pairings feed straight into Ray 3.2's keyframe inputs.

For disciplines where spatial or compositional judgement matters before rendering, this is the correct setup step. An interior designer composing a room reveal, an exhibition designer blocking a stand flythrough, or a graphic designer setting up a brand ident transition all benefit from seeing the first and last frames side by side before committing to a render. Boards makes that visual check fast and keeps it in one place.

Film Studio for Multi-Scene Sequences

The Film studio is built for cinematic sequences that span multiple shots. It sequences model calls across shots natively, which is significantly more efficient than chaining individual video nodes in Canvas when you are working with four or more scenes. For a pre-vis sequence with ten shots, Film handles the orchestration; Canvas would require ten manually wired nodes.

Use Film when the work is a linear sequence with a defined shot order. Use Canvas when the work is a complex pipeline with branching logic, parallel generation paths, or mixed output types feeding into each other.

Canvas for Complex Pipelines

Canvas gives you node-based control over the full generation pipeline. You can pipe the output of an image generation node directly into a video node, route that video output into an Assemble Film node for multi-shot batch orchestration, and connect any step to an LlmChatNode for prompt refinement mid-pipeline. This level of control is most useful for game developers building trailer sequences, product designers running parallel angle variations, or anyone whose video output depends on upstream generated assets rather than manually uploaded images.

Marketing Studio for Short Ad Video

For graphic designers and marketers building short ad videos, the Marketing studio generates social and ad creative and connects directly to Editing for caption overlays and final export. This keeps the pipeline compact: generate, caption, export, without routing through Canvas or Film for work that does not require multi-shot sequencing.

Credit Cost and When to Draft vs. Finalise

Ray 2 Flash costs fewer credits per render than Ray 3.2. That difference is the basis for every credit management decision in a video workflow. Run your first two or three iterations in Flash to validate motion, timing, and composition. Only then commit the higher credit cost of Ray 3.2 for the final output.

How strictly you need to follow this habit depends on your plan.

• Free tier: 150 one-time credits and one free video render. Enough to test one model before choosing a paid plan. The credits do not reset, so treat them as a trial of the workflow, not a production resource.
• Lite (£10/month): 1,000 credits and 1 concurrent generation. At this volume, the draft-then-finalise habit is not optional. One concurrent generation also means you are working sequentially, so plan your shot order before you start.
• Starter (£22/month): 2,500 credits and 2 concurrent generations. Enough to run a Flash draft and a Ray 3.2 final in parallel once you have validated direction.
• Pro (£42/month): 6,000 credits and 3 concurrent generations. Suited to professional workflows with multiple active projects.
• Studio (£84/month): 12,500 credits and 4 concurrent generations. At this level you can afford to run Flash and Ray 3.2 in parallel across multiple shots simultaneously, which meaningfully compresses turnaround time on multi-scene sequences.

Concurrency is often underweighted as a consideration. For video work specifically, the number of simultaneous renders your plan supports determines whether you can run a draft and a final in parallel or whether you must wait for one to complete before starting the next. On a Lite plan, that sequencing constraint is a real workflow limitation. On Studio, four concurrent generations means a five-shot sequence can be in progress across multiple models at once.

A practical discipline: before you start any multi-shot project, count the number of shots, multiply by the estimated renders per shot including drafts, and check whether that total fits within your monthly credit allowance. If it doesn't, decide upfront which shots warrant a Ray 3.2 final and which can stay at Flash quality.

Common Selection Mistakes and How to Avoid Them

Using Ray 3.2 for Every Draft

Ray 3.2 is the flagship model and the instinct is to default to it. But its credit cost exists precisely because it is doing more work per render. Ray 2 Flash was built so you do not burn high-fidelity credits on exploratory work. Using Ray 3.2 to validate timing or test a composition is a credit management mistake, not a quality decision.

Skipping Boards and Manually Uploading Keyframes

Manually uploading keyframe images one by one is slower and removes the visual check that makes Boards valuable. For disciplines like interior design, exhibition design, or automotive design, where spatial and compositional judgement is central to the work, seeing the start and end frames side by side before rendering is a meaningful quality step, not an administrative one. Boards was built for exactly this purpose.

Treating Video Generation as a Standalone Step

Prompting video cold, without a prepared keyframe, without upstream image generation, and without a defined output format, produces the weakest results across all twelve disciplines. The strongest outputs come from chaining image generation into video generation. Generate a hero still in Uni-1 or Scene Composer, use it as the first keyframe in Ray 3.2, and define the end state. That pipeline consistently outperforms text-only video prompts for any work where visual precision matters.

Ignoring the Film Studio for Multi-Shot Work

Manually stitching clips in Editing after generating them individually is the slowest approach to multi-scene video. The Film studio handles sequencing natively and sequences model calls across shots in a single workflow. For anything longer than a single scene, including pre-vis sequences, game trailers, product launch videos, or brand campaign content, Film saves significant time over assembling clips manually after the fact.

The simplest version of a good video workflow fits on one line: generate your key frames as stills, load them into Boards, set your first and last states, draft in Ray 2 Flash, finalise in Ray 3.2, sequence in Film. That pattern applies whether you are a product designer building a pitch video, a motion designer delivering a looping ident, or a game developer assembling a trailer pre-vis. The model names will change as the lineup evolves. The pipeline logic will not.