Hedra AI: Character 3 tutorial, pricing & Omnia comparison

Magic Hour publishes this guide and offers a separate talking-photo workflow. Hedra product, model, duration, resolution, and API claims are based on the linked official Hedra sources checked September 13, 2026; this is a documentation review rather than a retained output-quality benchmark.

Magic Hour publishes this guide and includes its own product in the comparison. Treat the recommendations as editorial guidance from a vendor, and verify the linked first-party product details and your own output requirements before choosing a tool.

What is Hedra Character 3?

Character 3 is Hedra's audio-driven character-video model. Its current model page lists a start image and required audio, with 540p, 720p, and 1080p output options. The older statement that Character 3 is limited to 720p is no longer consistent with that page.

Use it for a speaking or singing character when the source is a still image. Confirm the settings available in your account rather than assuming that every model on the platform has the same input, duration, or output limits.

Character 3 versus Hedra Omnia

Character 3 is Hedra's long-form audio-driven portrait model. It requires a start frame and audio, supports 540p, 720p and 1080p, and accepts up to 600 seconds of audio on the current API route. Start here when the deliverable is a presenter, singer, podcast clip or other speech-driven character performance.

Hedra Omnia launched in February 2026 as Hedra's more advanced model for character dialogue in dynamic environments with camera control. Evaluate it when the scene needs more than a largely portrait-led performance. Hedra currently presents Omnia and Character 3 as separate choices, so record the selected model instead of calling every Hedra output “Character 3.”

Hedra pricing: current monthly plans

Checked October 10, 2026: Hedra's pricing page lists the monthly plans above and commercial use on paid tiers. Ultra starts with 18,000 credits; its advertised higher allowance is not the base allocation. The video-generator overview explains supported workflows, but does not establish the current free-plan watermark or commercial-use conditions. Confirm those conditions and the selected model's charge in your account before client work.

Hedra Basic with monthly billing, captured October 10, 2026: $20 per month and 2,000 credits. This image shows Basic only; use the table for the other plans and confirm the selected model’s charge before buying. View official source

Do not turn plan credits into a promised number of videos without the selected model's current quote. Duration, output settings, and repeated attempts affect the amount of work your balance can cover.

Do Hedra credits roll over after renewal or cancellation?

Subscription credits reset each billing cycle; separately purchased credit packs do not expire. Hedra’s pricing FAQ says the monthly allowance is spent before pack credits. Size a recurring plan around work you expect to complete that cycle, rather than treating unused monthly credits as savings for a later project.

An unexpired balance does not guarantee continued access after cancelling. The same FAQ says most functionality becomes unavailable apart from limited watermarked generation, and you must resubscribe to spend remaining non-expiring credits. Before buying a large pack for occasional projects, account for the subscription needed when you return.

Character 3 API cost per finished attempt

The current Character 3 API page lists $0.025 per second at 540p, $0.05 at 720p, and $0.0625 at 1080p. A 60-second attempt is therefore $1.50, $3.00, or $3.75 before retries. A ten-minute 1080p attempt is $37.50 before retries. These are API rates for this model, not a conversion from Hedra subscription credits.

Budget by accepted output. If three one-minute 720p attempts are needed to approve one clip, the generation cost is $9 before any voice, editing, storage or review work. Save the model, resolution, audio duration, attempt count and actual charge with each result.

How to create the first talking video

1. Open Hedra and check the selected workflow

Sign in through Hedra, open Creative Studio, and choose the talking-character workflow and model. Interface labels can change; the required assets are the useful anchor: a starting image and an approved audio track.

2. Prepare one clear portrait

Use a well-lit face with unobstructed eyes and mouth. Leave room for movement around the head and chin. If you create a character image first, save the approved image instead of repeatedly prompting for an approximate replacement.

3. Approve the audio before animation

Record your line or generate speech, then listen to it by itself. Correct pronunciation, pauses, and the message before using it as an input. Hedra's photo-to-video guide describes this image-and-audio workflow.

Check the Character 3 API input limits

For the current Character 3 API route, the starting image must be no larger than 10.4 MB. Each audio file must run between 0.5 and 600 seconds and stay within 104.8 MB. These are API limits; check Studio separately.

For a thirteen-minute narration, plan separate clips of ten minutes and three minutes or shorter editorial segments. Reuse the approved portrait, cut at complete sentences, and assemble the returned clips in an editor. Do not truncate the audio merely to fit one request. This is a planning workflow, not a tested Hedra output.

4. Check the model, resolution, and charge

Choose settings appropriate to the deliverable and review the quoted credit use. Begin with a short segment that contains the expressions and pacing you need. A low-cost preview can screen the inputs, but it does not establish the quality of a different final mode.

5. Generate, review, and download

Watch the whole output at normal speed. Inspect the mouth at pauses, the eyes during movement, and the face shape through the sentence. Confirm the downloaded file's dimensions and export conditions before making a series.

Control which character speaks and how long the clip runs

For API projects with multiple people in the image, map each audio track to the intended speaker. The Character 3 API reference accepts up to four audio files. A single [x, y] point selects the speaker when one voice is supplied; multiple voices require one point per audio file, in the same order. These are API controls, not a promise that every Studio screen exposes them.

The optional duration uses milliseconds. Leave it unset to follow the supplied audio; setting it trims the audio or adds silence to reach that length. For example, an 18-second recording with duration 20,000 adds two seconds of silence; duration 12,000 cuts six seconds. These illustrate the documented rule, not tested outputs. Omit the override when you want the complete recording unchanged.

A sample brief for a product explainer

Use a licensed or original presenter image and a short line such as: “Here's how the cap opens. Turn it once, then lift.” Time the actual recording and make the instruction match the real product.

Pair the presenter with footage of the action. The talking character introduces the demonstration; it should not invent evidence that the product works. Do not present a generated actor as a real customer describing a purchase they never made.

Keep these three assets together: the approved image, the final voice recording, and the visible demonstration. When the instructions change, update the relevant asset and review the combined video. For longer structures, use the product video script templates.

Estimate the cost of a series

Suppose you need ten clips and each attempt receives a quote of Q credits. One attempt per clip costs 10Q; two attempts per clip cost 20Q. If only eight of those clips are usable, the cost per usable clip is the total spent divided by eight.

This is planning arithmetic, not a Hedra rate or an observed success rate. Record actual charges and usable outputs in your project. A subscription price alone does not show the cost of a finished campaign.

Troubleshoot before repeating a generation

Do not keep repeating an unchanged setup after the same failure. Isolate the image, audio, or setting responsible, or choose a different workflow.

Hedra vs Magic Hour vs HeyGen

Choose on your actual input, export, and workflow requirements. This guide does not establish that one provider has the most accurate lips, fastest processing, or lowest cost for every project. For a broader shortlist, see best AI talking-photo tools.

Choose Hedra for the character job it solves

Direct answer: Hedra belongs on the shortlist when audio-driven character performance is central. Compare it with broader talking-photo and video platforms when you also need generation, editing, multiple models or production handoffs.

Test. Use one portrait and 15-second script with pauses, emphasis and a mild head turn. Record model version, voice source, resolution, duration, queue and retries.

Review. Check identity, lips, teeth, blinking, head motion, framing and whether performance matches the audio. Longer clips reveal drift that a short showcase can hide.

Decision rule. Choose a specialist when character performance is the whole job. Choose a broader workspace when talking-photo generation is one step in a larger image, video and audio workflow.

Try the workflow: Open the matching Magic Hour tool. Use the same representative input and acceptance criteria before comparing results.