Early AI video generation gave you one thing: a clip. You described a scene, the model produced something, and you accepted whatever camera angle it chose. That constraint is now lifting.
Tellers added Runway Gen 4.5 in version 0.0.229 on April 6, alongside first/last frame control for LTX Video. Together, these features give you more directorial control over how AI-generated footage moves: which shot type you get, how a scene transitions, and where a clip begins and ends.
But the bigger shift is not just model capability. It is workflow.
Instead of prompting raw video models directly, you work with Tellers as an editing and generation agent. You can describe the result you want in natural language, and Tellers translates that intent into the technical prompt structure, motion language, and sequence of model calls needed to produce a coherent result. That makes camera direction more accessible for beginners, while still giving professionals the precision they expect.
This guide explains how that works in practice.
Runway Gen 4.5: Cinematic Prompting and World Consistency
Runway Gen 4.5 is now available in Tellers. Its defining characteristic is world consistency: characters, objects, and environments stay coherent across the duration of a clip without requiring explicit re-description. A character introduced in frame one can remain visually consistent in frame ten, even as the camera moves.
The model responds well to cinematographic language in prompts. Standard terms that work include:
- Camera movement:
slow dolly in,pan left,truck right,orbit around subject,crane up - Shot type:
close-up,medium shot,wide establishing shot,over-the-shoulder - Motion style:
handheld documentary style,smooth cinematic push,static locked-off frame
That matters because video generation quality often depends on knowing the right phrasing. Raw models tend to reward users who already know how to speak their language. Tellers helps close that gap.
If you describe a scene in plain English, Tellers can enrich and structure the request so the underlying model receives the kind of prompt it responds to best. If you already know the exact camera terminology you want, Tellers can follow those instructions faithfully and execute them across the underlying generation workflow.
Practical example
A user might ask for: “A founder in an office, make it feel cinematic and slowly move closer as they speak.”
Tellers can interpret that intent and translate it into the kind of language a model like Runway Gen 4.5 handles well, such as a slow push-in from a medium shot to a close-up with soft natural light and stable visual continuity.
That means you do not need to master every technical term to get a strong result. And as you work with Tellers, you also start picking up the vocabulary and prompting patterns that make AI video generation more reliable.
Tellers as the Layer Between Intent and Model Syntax
The main difference between using Tellers and prompting a raw GenAI video model directly is that Tellers acts as an agent, not just a text box.
With a raw model, the burden is on you to know:
- which model to use
- how to phrase the shot
- how to describe camera motion
- how to keep style and subject consistent
- how to break a scene into multiple generations
- how to bridge those generations into something editable
Tellers handles that orchestration for you.
You can give high-level creative instructions, and Tellers turns them into the technical lingo and generation steps required by the models underneath. It also helps maintain consistency across multiple clips in the same project, so you are not starting from scratch on every shot.
This is useful in two ways.
For users who do not know the technical vocabulary, Tellers lowers the barrier to entry. You can describe what you want naturally, and the agent helps convert that into something the model can execute well.
For advanced users, Tellers remains controllable. If you already know the precise language you want to use, you can specify shot type, motion, framing, and style directly. In that case, Tellers follows your instructions closely and orchestrates the many underlying model calls needed to turn a scene or video script into actual footage.
In other words, Tellers can both assist and obey:
- Assist when you want help translating intent into effective prompting
- Obey when you already know exactly how the shot should be directed
That dual behavior is important in real workflows. Beginners get better outputs without needing to become prompt engineers first. Professionals get speed, consistency, and less manual repetition.
Directing Camera Motion with Prompts
Tellers lets you direct camera behavior in natural language as part of the generation prompt. Instead of forcing you to think in isolated technical parameters, it lets you express subject, framing, and movement together, the way a filmmaker naturally would.
For example, you might ask for: “Wide establishing shot of a modern office at sunrise, slow dolly in toward the subject, smooth cinematic motion.”
Tellers can use that instruction directly, or refine it further so the selected model receives the most effective version of the request.
A practical workflow looks like this:
- Choose your model, such as Runway Gen 4.5
- Describe the scene and the intended movement in natural language
- Let Tellers translate and structure the generation request
- Review the clip and iterate if needed
- Combine the result with the rest of your footage in the Tellers timeline
Because Tellers sits above the models, it can help keep motion language and visual direction more consistent across multiple generations. That matters when you are building a sequence rather than just producing a single isolated shot.
It also makes Tellers a practical learning tool. By working with the agent, you gradually discover which terms, shot descriptions, and motion patterns produce the strongest results. Over time, you get better both at directing the AI and at understanding the craft language behind it.
LTX Video: First and Last Frame Control
The second feature worth knowing about is first/last frame control in LTX Video generation. This gives you keyframe-style control: you provide an image for the first frame, an image for the last frame, and the model generates the motion between them.
The applications are specific but useful.
Transition generation: You have a clip that ends on a particular frame, and a clip that starts on a different one. LTX can generate a transition that moves naturally between them, matching the visual content at both endpoints.
Controlled scene changes: For product shots or architectural videos, you can define the exact start and end framing, then let the model generate the movement in between.
Consistent visual bookends: If you are generating multiple clips in a sequence, shared first and last frames can help keep the visual flow coherent across segments.
When to use first/last frame control vs. prompt-based direction
Use first/last frame control when you have specific visual endpoints that must be respected, such as transitions between existing footage or clips that need to match incoming and outgoing frames precisely.
Use prompt-based direction when the visual endpoints are still open, but you know the kind of shot and movement you want.
The two approaches work well together: prompt-based direction for primary generated footage, and first/last frame control for transitions or tightly constrained shots.
Combining Both in a Real Workflow
A production workflow that takes advantage of both features might look like this:
- Primary clips: Generated with Runway Gen 4.5 from natural-language instructions, with Tellers translating the desired motion and shot style into model-ready prompts
- Transitions: LTX Video with first/last frame control, bridging between clips using outgoing and incoming frames as anchors
- Editing: All clips combined on the Tellers timeline, with generated footage mixed alongside uploaded material
- Export: Final cut exported directly, or handed off to a downstream edit
This approach is especially valuable when working from a broader scene brief or video script. Instead of manually prompting every shot one by one, you can let Tellers interpret the intent of the sequence, orchestrate the relevant model calls, and help keep the visual language coherent across the whole result.
FAQ
What camera movements can I specify in Tellers?
You can describe camera movement in natural language using terms such as dolly, pan, truck, orbit, crane, close-up, or wide establishing shot. Tellers can interpret those instructions and turn them into model-ready generations.
Do I need to know technical prompting language to get good results?
No. One of Tellers’ strengths is that it helps translate plain-language creative intent into the kind of prompt structure video models respond to best.
What if I already know the exact technical terms I want to use?
Tellers can follow precise instructions as well. Advanced users can specify shot language directly, and Tellers will execute that direction across the underlying generation workflow.
Is first/last frame control available for all video models on Tellers?
Currently, first/last frame control has been added to LTX Video generation. It is not available for all models.
Can I mix AI-generated footage with my own uploaded clips?
Yes. Tellers lets you bring in your own footage alongside AI-generated clips and edit them together on the timeline.
Which model should I use for cinematic camera work?
Runway Gen 4.5 is a strong choice for camera-directed generation because it responds well to cinematographic prompting and maintains visual consistency throughout the clip.
Camera motion has long been one of the hardest parts of AI video generation. The challenge is not only getting a model to pan when you ask it to pan, but getting a whole sequence of generations to feel intentional and coherent.
That is where Tellers changes the workflow. It does not just pass your prompt through. It helps translate creative intent into effective model instructions, preserves consistency across generations, and still gives professionals the control to direct shots precisely when they want to.
Start generating with Tellers or read the Tellers API docs to build video workflows programmatically.