A scripted presenter and an invented scene are different deliverables. If your brief calls for a keeper standing in a workshop with its own ambience and camera composition, first decide whether you need to animate an existing avatar or create the scene itself. This guide focuses on the latter HeyGen model workflow.
Published by Clout, a competing creative platform. We checked public documentation on October 24, 2026. The observatory brief, diagram and planner are original planning materials. We have not submitted these requests, generated a HeyGen clip or measured its quality or speed.
What HeyGen Video 1 does
The official model page presents HeyGen Video as scene generation with picture and sound, using text, image or reference inputs. Its developer guide identifies the model as heygen-video-1, built on MiniMax H3 and post-trained by HeyGen, and distinguishes it from avatar rendering.
For broader account, avatar and purchasing questions, see our HeyGen review and pricing guide. The workflow below focuses on scene requests and reviewing the returned clip. It does not imply that Clout currently exposes this model or that different platforms include equivalent generation credits.
Choose the mode from your input
| Mode | Input role | Review focus in our brief |
|---|---|---|
| Text to video | Describe a scene | Does the invented room and keeper match the brief |
| Image to video | Supply the opening image | Does the opening match and the later action remain coherent |
| Reference to video | Supply reference imagery or video | Which requested visual properties survived into the new shot |
A starting frame and a reference are different commitments. Our planner reflects this by using a first-frame asset placeholder for image mode and reference arrays for reference mode. Replace every placeholder with a real asset from your own workspace before considering an actual request.
Use the local scene request planner
Choose a mode, enhancement setting and whole-second duration. The planner fixes the draft to the 768p size class and a wide frame where the mode supports an explicit aspect ratio. In image mode, it omits the aspect-ratio field so the first-frame image governs the shape. The draft is available to inspect and download without providing an API key.
The current documented duration ceiling is 30 seconds for text or image mode with turbo or quality enhancement, and 15 seconds for reference mode or ultraquality or disabled enhancement. The planner rejects a duration outside the selected combination rather than quietly shortening it. It does not implement automatic duration selection, MCP behavior, every resolution or the complete API schema.
Reference mode requires an image or video. The planner checks up to nine images, three videos and three audio references, with twelve total. Download request plan JSON saves the draft and a note explaining its placeholders. These local checks are a planning aid, not proof that a future API request will be accepted.
Prepare the actual API handoff
The official create reference documents POST /v3/models/videos. Keep credentials on your own server, use your own uploaded reference IDs and consult the current schema before submission. Preserve the returned job ID and retrieve its terminal status before treating the requested work as a completed video.
Our downloadable examples are text mode, image mode and reference mode plans. They contain placeholder asset IDs and have not been sent. The planner source makes the local checks inspectable.
Direct one original observatory scene
The exercise is a fictional keeper beside an unmarked brass telescope. The keeper speaks one short line, the room has quiet ventilation and a distant mechanism clicks once. The acceptance rule is modest: a readable fixed composition, one coherent speaking action and no unintended music or lettering.
In an invented observatory workshop, a fictional keeper in a plain green coat stands beside an unmarked brass telescope. Fixed medium-wide composition. The keeper says: “The eastern lens is ready.” Quiet ventilation and one distant mechanical click. No music, lettering or logos. Keep hands still.
This is an untested original direction. Before using image mode, approve a still that already contains the intended opening composition. Before using reference mode, name the specific property each reference supplies: the keeper's appearance, telescope shape or room layout. A reference should have a clear role rather than being added simply because another input slot is available.
Review picture and sound as separate evidence
Watch the returned file with sound on, then again with sound off. Record whether the keeper's appearance and telescope proportions remain consistent. Inspect the beginning, speaking interval and ending rather than approving a single attractive frame.
Listen separately for the spoken line, ventilation and mechanical click. Note missing words, an extra voice, unwanted music or an effect that suggests an action absent from the picture. If the clip needs an edit, write the precise time and observation in the blank review ledger.
Keep requested and actual duration separate in your records. Preserve the exact prompt, settings, reference IDs, result file and real charge. Count accepted outputs against a delivery brief before calculating cost per usable shot. The planner does not estimate charges or validate the provider's billing.
Choose the production route for your project
For recurring original characters and scene exploration, compare the relevant workflow in Clout Image and Video or plan connected shots in AI Cinema. Select the platform from the actual input and deliverable you need, then run a small accepted-output pilot before expanding the sequence.
Use the matte-painting study when your real task is preserving a foreground while composing a distant environment. For a supplied 3D prop, the texture-painting study shows a native image-to-material workflow. Generating a scene, compositing a plate and editing a native asset solve different production needs.
