Direct Answer: Integrating Multiple Models into One Workflow
Short answer: RelayMe is a workflow builder where every video node picks its own model, so one prompt fans out to several models and the clips land side by side in a single task list — no switching between websites, accounts and prompt formats. As of September 2026 the video models live in the cloud workflow builder are Google Veo 3.1 and Veo 3.1 Fast, xAI Grok Imagine 1.5 and MiniMax-H3; Kling and Sora are not currently selectable, and the live list is always the one on the pricing page. Everything in this guide applies to any combination of the available models.
The practical payoff is model selection speed: the same shot description often produces very different camera work, texture and physics across models, and a side-by-side view settles it in seconds. Credits are billed per model tier, so testing on low tiers (no audio, short duration) keeps a full comparison round cheap.
The video models you can run in one workflow today, with prices
The public catalogue checked on 20 September 2026 lists Veo 3.1 Fast at US$0.15 per second with audio at 720p or 1080p, or US$0.10 without audio; Veo 3.1 is US$0.40 with audio or US$0.20 without. Grok Imagine 1.5 is US$0.15 per second with or without audio, with 480p, 720p and 1080p options. MiniMax-H3 lists a base rate of US$0.12 per second, but its selectable resolutions are 768P and 2K and its duration range is 5–15 seconds. Do not infer 4K or a 4-second mode from a price field alone; the supported options and the current task quote must agree.
For a supported 4-second, 720p comparison with audio, Veo 3.1 Fast costs US$0.60, Grok Imagine 1.5 US$0.60 and Veo 3.1 US$1.60: US$2.80 for those three generations. This is a calculation from the dated catalogue, not a measured benchmark. MiniMax-H3 cannot join a 4-second round because its minimum is 5 seconds. To include it, choose a duration shared by all candidates, such as 6 seconds, and record each model’s actual resolution rather than calling 720p and 768P identical.
Use price and supported settings to build a shortlist, then compare your own outputs. A model’s higher price does not establish higher quality for every shot, and a lower per-second rate does not establish a lower cost per usable clip. Keep the prompt, reference input and review criteria stable; record duration, resolution, audio, retries and actual task spend for each result.
Check the current pricing and credit rules before selecting a model; a dated example is not a current quote.
Standardizing Prompts for Multi-Model Comparison
Consistency is the key to meaningful comparison when using multiple AI models. Create a baseline prompt structure that includes the subject, motion description, and camera movement, and use this as the anchor for every node in your RelayMe workflow.
By feeding the same prompt into Veo 3.1, Grok Imagine and MiniMax-H3 simultaneously, you isolate the performance differences of the underlying models rather than the variance in your instruction set. Record these base prompt templates in your library for repeatable tests across different scenes.
Keep the baseline prompt short enough that every model interprets it cleanly, then add one variable element per test round, such as lighting or lens movement. Changing a single variable at a time is what makes the comparison conclusive rather than anecdotal.
Building the RelayMe Workflow Architecture
Begin by defining a trigger step in RelayMe that captures your initial creative idea. Once defined, map this input to concurrent branching paths where each path connects to a specific video generation engine—in this case, configured for your preferred models like Veo 3.1, Veo 3.1 Fast, Grok Imagine or MiniMax-H3.
Design your workflow to aggregate these outputs into a single review dashboard. This architecture allows for side-by-side analysis, helping you determine which model yields the most accurate animation or visual quality for your specific project requirements.
For recurring production, save the branching structure as a template so new prompts inherit the same three-way split automatically. Team members can then run standardized comparisons without needing to understand the underlying node configuration.
Before scaling a run, use the video cost breakdown to separate generation credits from human review and editing time.
Iterative Testing and History Management
RelayMe maintains a history of every task execution, which is vital for multi-model workflows. Use the task history to review previously generated videos, comparing how different versions of your prompts influenced the results from each individual model.
This documentation process serves as a learning loop. If one model produces a superior cinematic quality for a specific motion type, you can refine your workflow template to prioritize that model for similar future scenes, effectively building a customized 'best-in-class' production pipeline.
Tag or note each accepted clip with the model that produced it. Over a few weeks this record becomes your own evidence base for which engine suits product close-ups, scene transitions, or character motion, replacing guesswork with data.
Choosing the Right Workflow for Your Team
This workflow strategy is ideal for creators who require high-velocity prototyping and need to test multiple visual directions. It is not recommended for those who only need a single, final render without the need for model comparisons or structured version control.
Evidence suggests that centralizing operations in a single AI platform reduces context-switching friction. However, human evaluation remains essential to decide which model output is 'better,' as current AI systems do not inherently rank their own artistic quality objectively.
A reasonable starting point is to run multi-model comparisons only for new visual directions, then lock production batches onto the winning model. This keeps exploration costs bounded while preserving the quality benefits of comparison.
Run a limited comparison round when exploring a new visual direction, then use the candidate that meets your acceptance criteria for the production batch. Continuing to generate every shot on every model adds the costs of all branches; reserve that extra spend for questions the comparison can actually answer.
In the recorded 25 August 2026 production run, a 4-second 720p Veo 3.1 Fast clip with audio cost 600 credits and arrived in about one minute. That single observation is useful context, not a speed guarantee or a comparison of all current models. Reusing a prompt and image can reduce setup work, but each new model still needs its supported inputs and settings checked.