Money is a design constraint
The other units show how a scene is written, lit and voiced. This one shows what that scene costs, and whether a cheaper tool could make it without damaging what the learner sees.
ERUDITA is made with Claude directing HeyGen's cinematic Avatar Shots: one fixed avatar, one fixed voice, vertical 9:16, true Fellini black and white. HeyGen sells this on a credit plan: 5,000 credits for 240 euros. A single question started this unit: is there another tool that makes cinematic video, can be handled by Claude, and is cheaper or gives more credits?
The honest answer has two parts. Yes, several tools are cheaper per second. But for the shots that matter most to ERUDITA, speaking shots where the same character must keep the same face and the same voice across a whole series, the extra cost buys something the cheaper tools do not guarantee. This unit works through that trade-off with numbers.
Scope note. The first answer to the question was generic and did not use the project's own numbers. It was corrected once the project was named (see §14). The baseline in this unit comes from the real DU_06 production log: p0603 cost 78 credits for a 10.08 s clip.
What a learner can do after this unit
Written for production teams and learners on the REBOOT materials who need to plan a cinematic AI series on a real budget.
- Convert a plan price into a cost per clip and per second, and use it to budget a whole series, not a single scene.
- Compare tools on the right unit. Subscription credits, per-second API prices and flat per-clip prices cannot be compared until they are converted to the same clip.
- Separate what a tool generates from what it guarantees. A cheap model can make a beautiful shot and still fail to keep a character's face and voice.
- Choose which shots can move to a cheaper tool, and which must stay where the character lives.
- Design a fair test before changing a pipeline, and read its result against the cost saved.
- Explain the AI literacy point that the real expense of a minute of film is the work done before the first generation.
What HeyGen cinematic shots cost the project
Every comparison starts from one measured clip, not from a brochure.
| Item | Value | Source |
|---|---|---|
| Plan price | 5,000 credits for 240 EUR | Walter, 2 Oct 2026 |
| Cost of one 10 s clip | 78 credits | DU_06, p0603, Portrait, 720p, 10 s |
| Clips per 5,000 credits | about 64 | 5,000 / 78 = 64.1 |
| Cost per clip | about 3.75 EUR | 240 / 64.1 |
| Cost per second | about 0.37 EUR | 3.75 / 10 |
| One Seeker, Video 1 (7 clips) | 546 credits, about 26 EUR | p0601 to p0607 |
| One Seeker, Video 2 (6 clips) | 468 credits, about 22 EUR | p0611 to p0616 |
| One Seeker, both videos | 1,014 credits, about 49 EUR | 13 clips, no retakes |
| All six Seekers | 6,084 credits, about 292 EUR | 78 clips, no retakes, more than one 5,000 plan |
Two points follow. First, the series does not fit inside a single plan even if every clip works the first time, and no clip ever does: DU_04 and DU_06 both met colour drift, duplicated characters and wardrobe errors that forced retakes. Second, the silent clips (p0601 and p0604) cost the same 78 credits as speaking ones, although nobody speaks in them. They are the obvious first candidates to move.
Budget the whole series
Change the numbers to match your plan. The defaults are the project's own: 78 credits per clip and 240 euros for 5,000 credits.
Rough guide only. The moved-clip saving is in euros of HeyGen credits, the cheaper tool is in dollars and is not converted. It ignores your time spent checking and fixing the moved clips, which is the real cost.
What to try. Leave the defaults and read the euro figure: six Seekers at a 1.5x retake factor need about 9,100 credits, nearly two plans. Then set "silent clips moved" to 2. You save only about 15 percent, because two of every thirteen clips are silent. A saving of that size is real, but it is not the transformation a price comparison suggests.
Who else makes cinematic shots
Prices converted to one 10-second clip so they line up with the baseline. All figures are approximate US dollars, taken from comparison sites in September and October 2026.
| Tool | How it is billed | About per 10 s clip | Strength | Limit for ERUDITA |
|---|---|---|---|---|
| HeyGen cinematic (baseline) | Credits, 78 per clip | about 4 USD (3.75 EUR) | Fixed avatar, fixed voice, lip-sync | Credit blocks and retakes |
| HeyGen API, Cinematic Avatar | Prepaid wallet, flat per clip | 7 USD | Same product, callable from code | More expensive than the web plan |
| Kling 3.0 | Subscription credits, or API per second | about 1 USD (API); plan from about 7 USD a month | Cheapest, native 4K, multilingual audio, daily free credits | Weaker consistency across longer sequences |
| Veo 3.1 Lite | API per second | about 0.50 USD | Very cheap silent footage | No audio on this tier |
| Veo 3.1 Fast | API per second | about 1.50 to 2 USD | Synchronised dialogue and sound | No fixed character voice between clips |
| Veo 3.1 Standard | API per second | about 4 USD | Best cinematic lighting and audio of the group | Same price as HeyGen |
| Runway Gen-4.5 | Subscription credits, 12 per second | about 2 to 3 USD | Film-style lighting, scene consistency, editing tools | No avatar and voice system |
| Higgsfield | One subscription for many models | depends on the model used | 30+ models behind one Claude connector, cinematic camera presets | Generates only; check its current plans |
| fal.ai | Pay as you go per model | about 1 to 4 USD by model | Compare models from one place, no subscription | Needs an API key; raw infrastructure |
| Sora 2 | Closed | not available | Consumer apps closed in April 2026; the API was scheduled to retire on 24 September 2026 |
How the Runway number was reached. Standard is 15 dollars for 625 credits and Pro is 35 dollars for 2,250 credits. At 12 credits per second, Standard gives about 52 seconds and Pro about 187 seconds. That is roughly 0.29 and 0.19 dollars per second, or 2.9 and 1.9 dollars per 10 seconds. The same sum works for any credit plan: price / (credits / credits per second) x clip length.
Why the table is not a ranking. The cheapest row is Veo Lite at about half a dollar, but it has no sound. The strongest-looking row for lighting may be Veo Standard, but it costs as much as the tool it would replace. A cost table only becomes useful once it is read against what each shot needs.
What if the avatar only talks?
Cinematic Avatar Shots are the expensive part of the series. If a shot is only a face and a voice speaking to camera, a standard avatar video uses far fewer credits. This section prices that option and says what it gives up.
If you use only talking avatars, the same credits go a very long way. These figures use the rates on HeyGen's own help pages: Avatar IV with a photo look at 16 credits per minute, Avatar IV with a video look at 31, Avatar V at 48, and Presenter mode at 0.3 credits per second. Avatar III at 3 credits per minute comes from third-party reviews. I haven't confirmed any of them in your account.
Revised at 14:12 on 2 October. The first version of this section used 20 credits per minute for Avatar IV/V, taken from third-party reviews. HeyGen's help pages give different rates for each engine, so the table below was recalculated. Sofia was built from a single portrait, so her row is the Avatar IV photo look.
| 600-credit plan | 5,000 credits (240 €) | |
|---|---|---|
| Avatar IV, photo look (16 credits per minute): minutes | 37 | 312 |
| Avatar IV, photo look: 10-second clips | 225 | 1,875 |
| Avatar IV, video look (31 per minute): minutes | 19 | 161 |
| Avatar V (48 per minute): minutes | 12 | 104 |
| Presenter mode (0.3 per second): minutes | 33 | 277 |
| Avatar III (3 per minute, from reviews): minutes | 200 | about 1,666 (27 hours) |
| Cinematic shots, for comparison (78 credits per 10 s) | 7 clips | 64 clips |
On the 240 € plan, Avatar IV with a photo look works out at about 0.77 € per minute, or 0.13 € per 10-second clip. Your cinematic clips cost about 3.75 € each, so a talking-head clip is roughly 29 times cheaper per second.
Applied to DU_06, the 13 clips per Seeker are 130 seconds, which is about 35 credits as Avatar IV photo clips instead of 1,014 as cinematic ones. All six Seekers come to about 210 credits, or about 310 with a 1.5x retake factor. That fits inside a single 600-credit plan.
Two caveats:
- Credit rates: HeyGen changed its credit model in May and June 2026, and reviews and help pages give different figures. Make one standard avatar clip in your account and read the credits it uses before planning around these numbers.
- The 600-credit plan: Some reviews list 600 credits under Creator at $29, and Pro at 1,000 credits for $49. Check which plan you mean and what it costs before comparing it with the 240 € plan.
A talking head is also not the same product. You would lose the church interior, the camera push-in, the lighting and the Fellini black-and-white look. HeyGen would only need to render the face and voice. One option is to use avatar clips for the dialogue-heavy Video 2 requests and keep cinematic shots for the key moments, but that would change the visual language of the series.
Defaults: 600 credits at 16 credits per minute (Avatar IV, photo look). Enter 5000 credits and 240 EUR to see the larger plan, or 48 for Avatar V. Replace the rate with the number your account shows for one standard avatar clip.
What the avatar-only route gives up
- The place. No church interior, no street in Peniche, no window light. The avatar speaks in front of a background.
- The camera. No slow push-in, no change of angle, no cut within a shot.
- The Fellini look. Black and white can be applied in post, but the staging, shadow and film grain of a cinematic shot cannot.
- The performance beat. Silent moments such as p0601 and p0604 have no talking head to render, so they do not belong in this mode at all.
Where it could fit
Direct address to camera with no setting that matters to the story.
- Some Video 2 request shots, if the look can be matched
- Explainer, welcome and instruction clips
- Language versions of a finished scene
Anything where place, light or camera carries the meaning.
- The reading and reaction shots of Video 1
- Shame, anger and revelation beats
- Any shot cut next to a cinematic shot in the same scene
Avatar III, IV, V and Cinematic: what each can and cannot do
Three separate choices sit behind every HeyGen clip: what the avatar is built from (a photo, a video or a drawn character), which engine moves it (Avatar III, IV or V), and which mode you generate in (Presenter or Cinematic).
| Engine or mode | Built from | What it does well | What it cannot do | Credit rate (HeyGen help pages) |
|---|---|---|---|---|
| Avatar III | Photo | Precise lip-sync, up to 4K except on photo avatars. The cheapest. | No motion prompts, basic quality | Not stated; reviews say about 3 per minute |
| Avatar IV | Photo or video, including drawn or stylised characters | More facial expression and head movement, accepts motion prompts, works with non-human characters | No 4K | 16 per minute (photo look), 31 (video look) |
| Avatar V | Video only: about 15 seconds of a real person, a "digital twin" | Best identity match, upper-body gestures learned from your footage | Cannot be used with photo avatars, public expressive avatars or legacy studio avatars | 48 per minute |
| Presenter mode | Any supported avatar | Talking head facing the camera, up to 30 minutes, gestures chosen through a motion prompt | No change of setting, no camera movement | 0.3 per second (about 18 per minute) |
| Cinematic mode | Any avatar | The avatar acts inside a scene, with 360° camera control and prompts for action, camera and atmosphere | Maximum 15 seconds per clip, so no long speech | Help page says a flat 60 per generation; the project paid 78 for 10 s |
Which avatar type goes with which engine. A photo avatar works with Avatar III and IV, not V. A video avatar (a digital twin of a real person) defaults to Avatar V. A virtual character, an AI-generated or stylised figure, works with Avatar IV, and HeyGen's pages disagree on whether it can also use Avatar V. Check what your account offers for Sofia before relying on either answer.
What this means for ERUDITA
Shots where place, light and camera carry the meaning, as they do now.
- The reading and reaction shots of Video 1
- Shame, anger and revelation beats
Plain direct address. Sofia was built from a single portrait, so she is a photo-based character and this is her option.
- Some Video 2 request shots, if the look can be matched
- Explainer and instruction clips
It needs a video of a real presenter, so it suits a person such as your own Eduagility avatar, not a fictional Seeker.
- Project introductions presented by a team member
No motion prompts and the weakest expression. The low price does not make up for it in a series that relies on performance.
Connectors generate; assembly stays in the workspace
The original question asked for software "that Claude can handle the editing" of. The distinction between generating and editing matters here.
| Connector | Claude can | Claude cannot |
|---|---|---|
| Higgsfield MCP | Generate image, video and audio from 30+ models, run multi-step creative workflows, poll for finished jobs | Cut, trim or assemble a timeline |
| Runway MCP | Submit and monitor Gen-4.5 (and other model) jobs; assets appear in your Runway library | Replace a full editor |
| Kling MCP | Generate, extend and lip-sync clips on your existing Kling plan | Hold a character across a series by itself |
| fal.ai MCP | Read prices and schemas, then run Kling, Luma and other models from one place | Edit; needs your API key |
| Claude's own workspace | Assemble clips with ffmpeg: crossfades, desaturation, loudness, subtitles, voice-over mixing, duration checks | Make new moving picture by itself |
| HyperFrames (HeyGen) | Compose and render videos from structured projects | Replace cinematic generation |
In practice the editing already happens in Claude's workspace, as in DU_06: desaturating p0603 in post, building the voice-over and the breath separately, assembling with frame-exact crossfades. Adding a second generator does not change that part. It only changes which clips arrive.
Connectors that sign in with an account use that account's plan, so a connector never makes a tool cheaper. It only makes it reachable from the conversation. Connectors must be added by the account owner in Claude's connector settings, and each one needs its own sign-in.
Which shots go where
Sort each Plano by what it must deliver, then price it. Speaking shots stay with the avatar; shots without a speaker can be moved.
| Plano | Content | Speaker | Suggested tool | Reason |
|---|---|---|---|---|
| p0601 | Sofia reads the envelope in silence; voice-over added in post | None on screen | Candidate to move | No lip-sync needed; face must still match |
| p0602 | Reads her article aloud | Sofia | HeyGen | Fixed voice, delivery from punctuation |
| p0603 | Reads the register, clinical | Sofia | HeyGen | Already generated and delivered |
| p0604 | Silent beat, breath added in post | None | Candidate to move | Face and light only |
| p0605 to p0607 | Shame, anger, direct address | Sofia | HeyGen | Performance carries the scene |
| p0611 to p0616 | Video 2, the request, all to camera | Sofia | HeyGen | Voice and face must match Video 1 |
The arithmetic. Two clips out of thirteen is 15.4 percent. Moving them saves 156 credits per Seeker, about 7.50 euros, or about 45 euros across six Seekers, minus roughly 2 dollars a clip on the cheaper tool. Saving beyond that means moving speaking shots, which is where consistency risk starts.
A correction to an earlier estimate. The first answer in the conversation said a hybrid could cut spend by 20 to 30 percent. With the real clip breakdown, moving only the silent clips saves about 15 percent. Larger savings would need establishing shots that are not in the current script.
CRITICAL VISUAL RULE: Fellini black-and-white film, true monochrome only. No color of any kind, no sepia, no warm tint, no cool tint, no green. Vertical 9:16.
Interior of an old stone church, empty wooden pews. A cloud passes over a high window: the shaft of light fades and the church turns grey and soft. A woman in her thirties with long dark wavy hair, blazer, silk scarf and a small pendant necklace, sits alone. Slow push-in to a tight close-up of her face. She lays an open magazine beside a folded document. Her hand stops on the page. She is processing, not crying. The shot is completely silent.
(Use the same words as the HeyGen prompt, so the two models can be compared on the same scene.)
Do not change the pipeline on a brochure
A fair test of one silent clip costs a few dollars and settles the question with evidence from your own series.
- Choose p0604, because it is silent and has the simplest brief.
- Generate it in HeyGen with the existing prompt (78 credits), and on one cheaper model with the same text and the Sofia reference image.
- Run the series QA checks on both: true first and last frame, contact sheet, per-frame channel difference for grayscale, full-resolution scan for a duplicated Sofia.
- Compare the face, wardrobe and light against p0603 side by side, not against memory.
- Record the cost, the time spent fixing the clip, and what had to be corrected in post.
- Decide with a rule written before the test (see §08), not after seeing the result.
| Criterion | Pass if | Measure |
|---|---|---|
| Likeness | The woman is clearly the same Sofia as in p0603 | Side-by-side contact sheet; a second viewer, not only the author |
| Monochrome | True grayscale after desaturation, no tint | Channel-difference check per frame |
| Wardrobe | Blazer, scarf, pendant; no pearls expected, no green | Contact sheet |
| Single character | Exactly one Sofia in every frame | Full-resolution frames at start, middle and end |
| Silence | No speech or unwanted audio | silencedetect, not a transcript |
| Net cost | Saved credits exceed cost plus extra checking time | Euros and minutes |
Write the rule before the result
Decide in advance what would change your mind. Otherwise a low price will always look like a good answer.
The clip has a speaker, a fixed voice, or a close-up that must match another clip exactly. Also when a retake would cost more of your time than the credits saved.
- All dialogue shots, Video 1 and Video 2
- Any clip that follows or precedes a dialogue shot in the same scene
The clip has no speaker, the face is small or in shadow, and the test passed on likeness and monochrome.
- Silent beats such as p0601 and p0604
- Establishing shots of Peniche, if later added
- Look-development tests and camera tests before the final take
The six-Seeker budget at your real retake rate needs more than one plan a month, and the credits do not roll over. Compare the second plan with the saved clips.
- Check the real retake rate after the first three Seekers
- Re-read the credit rules on the account page
A tool or plan promises "unlimited" generation, a credit window disagrees with support, or a connector asks for a purchase.
- Never accept an upgrade window on the owner's behalf
- Record the exact on-screen text for support
Met on the way
Problems from the cost question itself and from the DU_06 production that supplies the baseline.
| Problem | What happened | Status or fix |
|---|---|---|
| Question read as generic | The first reply compared AI video tools in general, not the ERUDITA cinematic shots. | Corrected once the project was named; baseline rebuilt from the DU_06 log. |
| Overstated saving | An estimate of 20 to 30 percent was given before counting silent clips. | Recomputed: about 15 percent for silent clips only. |
| Prices in different units | Credits, dollars per second and flat per-clip prices do not compare directly. | All converted to one 10 s clip, with the formula shown. |
| Stale pricing pages | Reviews of HeyGen plans disagree on credit counts; HeyGen changed its credit model in May and June 2026. | Use the account page and the project's own numbers. |
| Cinematic generation blocked | "Unlock more Avatar Shots" appeared while support said rollover credits were usable. | Sent to HeyGen's technical team; cleared before p0603, returned for p0601. |
| Colour leak | p0603 kept a green tint on the blouse. | Full desaturation in post; later prompts name the blouse as grey and add "no green". |
| Pearls ignored | HeyGen followed the avatar image's pendant. | Accepted; a second look is possible. |
For anyone budgeting AI film
- Price the series, not the scene. A scene that fits a plan can be one seventh of what the project needs.
- Measure your own clip. One real credit figure beats five review sites.
- Count retakes in advance. They, not the first take, decide the bill.
- Silent shots are the cheap win, and a small one. They are the easiest to move and a minority of the clips.
- Consistency is the product. In a series with six comparable characters, a cheaper face that drifts costs more in checking than it saves.
- A connector is a doorway, not a discount. It uses your plan and your sign-in.
- Check a claim against the project before answering it. A comparison that ignores what is actually being made can be accurate and still unhelpful.
What actually goes into two minutes
Written for the REBOOT team as a preface to the case study that follows, reproduced here in full, the same text every unit in this series carries in this slot.
There is a fantasy spreading across social media: that you can simply talk to AI and it will produce whatever you imagine. They named it prompting, JSON, context.
It won't.
AI can create the impression that it is talking to you, understanding you, even collaborating with you. But much of that feeling is human projection. Generative AI does not understand in the human sense. It works through extraordinarily sophisticated mathematical models that predict and generate patterns from what they have learned from vast amounts of data.
And that is precisely what makes this moment so extraordinary.
The paradox is that the more powerful AI becomes, the more human literacy matters.
Producing a two-minute cinematic clip is not simply a matter of writing a prompt. You need to understand cinema: framing, camera movement, rhythm, lighting, performance, and editing. You need storytelling, literature, and plot. You may need philosophy to construct the dilemma, design to create the visual language, typography to establish identity, sound to build atmosphere, and psychology to understand how a character should behave.
And before generating a single useful scene, there may already be days of work behind it: creating the characters, defining their profiles, maintaining visual consistency, designing their relationships, placing them within the story, building the plot, preparing locations, establishing the cinematic language, and deciding what information the audience should, and should not, know.
Then comes the AI.
A two-minute clip may take two minutes to watch. It can take days to create.
The real skill is not learning how to ask AI for things. It is learning what needs to exist before you ask.
So start by understanding the tools. Learn the difference between an LLM and multimodal AI. Learn what each can and cannot do. And, above all, develop the cultural, creative, and critical literacy required to direct them.
AI does not eliminate expertise.
It makes expertise visible.
Before changing a tool in a series
- Frame 0 and the last frame, literally, of every delivered clip.
- Contact sheet of every clip, compared with an approved clip of the same character.
- Measure grayscale, do not eyeball it: channel difference per frame.
- Scan full-resolution frames for a duplicated character at start, middle and end.
- Check silence with silencedetect, not with a transcript.
- Transcribe every spoken line and compare it with the intended text.
- Repeat the environment and wardrobe paragraphs word for word in every prompt of a scene, on every tool.
- Confirm the settings before every Generate: only the right avatar attached, Portrait, 720p, 10 s, Enhance prompt off.
- Record cost and checking time for each clip so the next budget uses real numbers.
Resolved so far
- Baseline: 78 credits per 10 s clip, about 3.75 EUR, about 49 EUR per Seeker without retakes.
- Alternatives mapped: Kling, Veo, Runway, Higgsfield, fal.ai, with Sora 2 ruled out.
- Saving estimate corrected: about 15 percent for silent clips only.
Still open
- Test clip: p0604 on one cheaper model against HeyGen, on the §07 criteria. Not yet run.
- Connector: add Higgsfield or Runway to Claude only if the test is worth running; the owner adds it in settings.
- Real retake rate: measure it on the first three Seekers before choosing a plan size.
- Plan terms: confirm on HeyGen's account page how credits renew and roll over, since support and the app window disagreed on dates.
- Prices: re-check every competitor figure before any purchase.
- Avatar credit rate: make one standard avatar clip, read the credits it uses, and compare it with the 16 credits per minute used in §04b.
- Plan names: confirm which HeyGen plan has 600 credits and its price, since reviews list that amount under different plans.
- D-ID: check its own pricing page and whether it can be reached from Claude, before deciding whether it deserves a test.
- Unit number: assign it in the library ledger, and add the link once published.
How this unit began
The exchange of 2 October 2026 (Europe/Madrid time) that led to this unit. Walter's messages are quoted exactly as written. Claude's replies are condensed to their substance.
"We use Claude, linked to Hey Gen, to produce cinematic videos. For 5000 credit they charge 240€. Is there any other software to produce cinematic videos that Claude can handle the editing that is cheaper or gives more credits?"
Searched current pricing for AI video generators, HeyGen's plans and API, and Claude connectors for video (Higgsfield, Runway, Kling, fal.ai).
Answered in general terms: several tools are cheaper per second (Kling 3.0, Veo 3.1 tiers, Runway Gen-4.5, Higgsfield as a hub, fal.ai pay as you go), HeyGen's Cinematic Avatar is a flat 7 dollars per clip on the API, and the connectors generate clips but do not edit them.
"I am talking about the cinematic videos we do for Erudita, are you aware of what I am talking about"
Attached: the DU_06 Sofia dossier.
Confirmed the project: HeyGen Single scene, Cinematic, vertical 9:16, 10 s, 720p, Fellini black and white, p0603 at 78 credits. Rebuilt the numbers from that baseline (about 64 clips, 3.75 euros per clip, about 49 euros per Seeker) and argued that for speaking shots the fixed face and voice justify HeyGen's price, while silent clips could move. Offered to set up a test of one silent clip.
"Create an HMTL Erudita style with all this information as an educational Unit"
Built this dossier on the series layout. While assembling the figures, the earlier "20 to 30 percent" saving was recomputed from the clip breakdown and corrected to about 15 percent, and the correction was recorded in §06 and §09.
"https://www.d-id.com/ is not in your table why?"
Read D-ID's site. It makes talking-head avatars and real-time agents, not cinematic scenes, so it was left out of the alternatives. Its homepage shows no prices and third-party figures disagree, so no number was given.
"and if we do only avatars talking meaning how many can I produce with the pro plan 600 credits and with the 240€ pro plan"
Priced the avatar-only mode: at 20 credits per minute, 600 credits give 30 minutes and 5,000 credits give 250 minutes. Warned that the credit rates come from third-party reviews and that the plan names and credit counts need checking in the account.
"add it to the Unit"
Added §04b "Avatar only" with the comparison table, a small calculator, what the mode gives up, where it could fit, and the D-ID note. This conversation record was extended with the new turns.
"did you included exactly this answer?"
Said no: the figures and caveats were all there, but reorganized, with extra material added.
"I htink should be clear stated" (the 13:59 answer, with its table), "The table is important"
Replaced the summary in §04b with the 13:59 answer as written, table included, placed at the top of the section.
"what is the difference bewteen the different Avatars version of Hey Gen what can I do and what i can't do and for Erudita which one I could choose."
Read HeyGen's own help and developer pages. Explained the three layers (avatar type, engine, mode), what each engine can and cannot do, and which to use for ERUDITA. Found that HeyGen's pages give 16, 31 and 48 credits per minute, not the 20 quoted from reviews, and said §04b needed correcting.
"can you add this to the Unit?"
Added §04c "Avatar versions" and recalculated the §04b table and calculator with HeyGen's own rates. The table in §04b is therefore no longer the 13:59 text word for word, and the change is marked in the section.
Where the numbers came from
Third-party comparison pages and vendor help pages, read in the first days of October 2026. Re-check before buying.
- Best AI Video Generator 2026: Veo vs Kling vs Gemini Omni, per-minute API prices and Runway credit rate.
- Runway vs Kling AI: 2026 Pricing Comparison, plan prices and credits.
- Veo 3 vs Kling 3 vs Runway (2026), Veo Lite and Standard per-second prices, Kling and Runway plans.
- HeyGen API Pricing Explained, pay-as-you-go model and separation from subscriptions.
- HeyGen API pricing explained (2026), flat Cinematic Avatar price and per-minute rates.
- Best MCP Servers for AI Video Generation (2026), connector list and endpoints.
- Runway brings AI video and image generation into Claude, Runway MCP.
- How to Turn Claude Into a Full Creative Studio With Higgsfield, Higgsfield MCP.
- Best AI Video Generators 2026, Sora shutdown dates and free tiers.
- HeyGen Review 2026: Pricing and Features, credit rates per minute for Avatar III, IV/V and Video Agent.
- HeyGen Pricing 2026: Plans, Credits and The Real Cost Explained, plan credit counts.
- Avatar V is now available on HeyGen, engine differences and credits per minute.
- Single Scene tool, Presenter and Cinematic modes, limits and credits.
- Avatar and Voice FAQ, avatar types and which engines they support.
- HeyGen Documentation: Models and HeyGen Avatar Lab, engine capabilities and limits.
- D-ID, product description. Pricing from fluxnote.io and costbench.com, which disagree.
- Project figures (5,000 credits for 240 EUR; 78 credits per clip): Walter's account and the DU_06 production log, 1 and 2 October 2026.