Happy Horse is Alibaba's video model, and on 8frame it is the one that takes several character images at once: a 5-second clip costs 95 credits at 720p (about $0.95) or 189 at 1080p, whether you start from a prompt, one start image, or 1 to 9 reference images. Kling v3 Standard costs 86 credits for 5 seconds with sound (57 silent), but on 8frame it animates one start image and takes no reference images; Kling's reference mode is a separate model, Kling O1 Reference, at 73 credits for 5 seconds with up to 7 inputs. So with up to 7 references in a 5- or 10-second clip, Kling O1 Reference is the cheaper run; with 8 or 9 characters, or at any other length from 3 to 15 seconds, only Happy Horse does it.
TL;DR
- Happy Horse: 95 credits per 5 s at 720p, 189 at 1080p (the node default); 3 to 15 s; prompt, start image or 1 to 9 references at one price
- Kling v3 Standard: 86 per 5 s with sound, 57 silent, 171 for 10 s; 3 to 15 s; start image required, no references on 8frame
- Kling O1 Reference: 73 per 5 s, 146 per 10 s; up to 7 reference images or elements; 5 or 10 s
- Resolution moves the price more than the model does: Happy Horse at 1080p is twice its 720p price; Kling v3 Standard files on 8frame are 720p
What each model is, in the vendors' words
Happy Horse. Alibaba Cloud's launch post of April 28, 2026 calls HappyHorse 1.0 "a video generation model designed to help creators produce high-quality, cinematic-style video content", "Developed by Alibaba Token Hub (ATH) Business Unit", with "Text-to-Video (T2V), Image-to-Video (I2V), and Subject-to-Video (S2V) generation" and "up to 15 seconds of 1080p video with multiple shots". fal, which serves it to 8frame, sums it up as "Joint audio-video generation in a single pass" with lip-sync in seven languages. Of the 1.1 release (June 23, 2026) Alibaba says it "significantly improves visual consistency between reference images and generated outputs in Reference-to-Video (R2V) tasks": Alibaba's claim, not our measurement.
Kling v3. Kling's VIDEO 3.0 guide: the model "generates up to 15 seconds of continuous video, with a flexible duration ranging from 3 to 15 seconds" at "1080p and 720p", with "native audio-visual output". fal's v3 Standard image-to-video endpoint, the one 8frame calls, marks the start image as required.
Kling O1. Kling's VIDEO O1 guide: "Supports uploading 1-7 reference images or elements in the input area", "generations anywhere between 3-10s", "two modes: 1080p and 720p". fal's O1 reference-to-video endpoint bills "$0.112 per second" and caps "Maximum Inputs: 7 total".
The nuance: "reference-to-video" is three different things
Happy Horse counts subjects. fal's endpoint takes "1-9 reference images" and you "Reference subjects as character1-character9 matching the order of image_urls". Alibaba's own Model Studio API (happyhorse-1.1-r2v) gives the same count, "Number of reference images: 1 to 9", tagged [Image 1], [Image 2] in the prompt. The count does not change the price: 8frame bills Happy Horse by seconds and resolution only.
Kling O1 counts inputs. Kling allows "1-7 reference images or elements", and an element is itself a set of stills: "You can upload/generate multiple images from different angles (up to 4 images) to form an element." fal's cap of "7 total" is "Combined: elements + reference images + start frame", so seven inputs can be seven people or two people from several angles. On 8frame, O1 Reference takes up to 7 reference images.
Kling v3 on 8frame counts one. The v3 endpoints 8frame calls are image-to-video with the start image required, and 8frame sends no element references on v3, so the start frame is the only identity the model gets: that is why characters drift between Kling v3 clips that begin from different stills.
One more gap: Alibaba's own API also offers 480P and nine aspect ratios for reference-to-video; fal, and therefore 8frame, offers 720p and 1080p and five ratios (16:9, 9:16, 1:1, 4:3, 3:4). There is no 480p Happy Horse on 8frame.
Price table
fal list prices as shown on 2026-10-18; 8frame credits as the app charges them, a credit about $0.01 at pack rate. 8frame's credits follow the provider's per-second rate: Happy Horse at 720p for 5 seconds is 95 credits, at 1080p 189.
| Model, as run on 8frame | fal price per second | 5 s | 10 s | Reference inputs | Lengths |
|---|---|---|---|---|---|
| Happy Horse, 720p | $0.14 | 95 | 189 | 1 to 9 images, or one start image | 3 to 15 s |
| Happy Horse, 1080p (node default) | $0.28 | 189 | 378 | 1 to 9 images, or one start image | 3 to 15 s |
| Happy Horse video-edit, 720p / 1080p | $0.28 / $0.56 | 189 / 378 | 378 / 756 | input clip plus up to 5 images | billed on the clip, capped at 15 s |
| Kling v3 Standard, sound on / silent | $0.126 / $0.084 | 86 / 57 | 171 / 114 | one start image | 3 to 15 s |
| Kling O1 Reference | $0.112 | 73 | 146 | up to 7 reference images or elements | 5 or 10 s |
fal's video-edit page shows two rates ($0.14 and $0.28 in its pricing table, $0.28 and $0.56 in its playground note); 8frame's table uses the higher pair and bills on the input clip's length, capped at 15 seconds, so a 60-second input (the maximum) bills as 15: 567 at 720p, 1134 at 1080p.
How it works on 8frame
Happy Horse is one node, and the mode follows what you attach: a video makes it a video edit, one or more reference images make it reference-to-video, a single start image makes it image-to-video, a prompt alone makes it text-to-video. With a start image and references both attached, the references are sent and the start image is dropped. The node opens on 5 seconds, 1080p, 16:9 (189 credits); switch to 720p before you generate and the same clip is 95. Every Happy Horse generation file we checked carries an audio track (100 percent in the 120 days to 2026-10-02); there is no sound toggle for generation, only the video-edit switch that keeps or replaces the original track. An audio track is not proof of useful sound, so listen before you ship.
Kling is one node whose picker opens on Kling v3 Standard, 5 seconds, sound on: 86 credits. v3 Standard and Pro need a start image, run 3 to 15 seconds, and sound is a toggle (Pro: 76 silent, 114 with sound). Measured Kling v3 Standard files are 720p. For references, switch the picker to O1 Reference: up to 7 images, 73 credits for 5 seconds and 146 for 10, no sound tier in its price. How to use Kling v3 Pro covers when Pro pays off; Kling v3 vs Seedance 2.5 covers the other reference-capable model weighed against Kling.
Which one for which job
- Several named characters from stills: Kling O1 Reference (73 for 5 s) for up to 7 inputs at 5 or 10 seconds; Happy Horse at 720p (95) for 8 or 9 references or any other length.
- One approved frame that has to move: Kling v3 Standard, 57 silent or 86 with sound; Happy Horse image-to-video at 720p is 95.
- Editing a clip you have: Happy Horse video-edit at 720p, 189 for a 5-second input, up to 5 reference images; editing existing video with AI lists the other routes.
We have not measured which model keeps a character's identity better; rely on the counts and prices above. Reference image limits by model covers the other video and image models.
FAQ
What is the Happy Horse AI video model?
Alibaba's video generation model, built by its Alibaba Token Hub business unit and launched in April 2026 (version 1.1 on June 23, 2026): text-to-video, image-to-video and subject-to-video generation plus video editing, up to 15 seconds of 1080p with synchronized audio. On 8frame it runs through fal as one node with those four modes.
Happy Horse vs Kling: which is cheaper for a 5-second clip?
With a start image: Kling v3 Standard, 57 silent or 86 with sound, against 95 for Happy Horse at 720p. With references: Kling O1 Reference, 73, against 95. At 1080p Happy Horse is 189.
How many character references does Happy Horse take?
1 to 9 reference images, one subject each, referenced in upload order; the count does not change the price. In video-edit mode the cap is 5.
Can Kling v3 use character references on 8frame?
No. Kling v3 Standard and Pro are image-to-video on 8frame: one start image, required. For references, use Kling O1 Reference (up to 7 inputs, 73 credits for 5 seconds) or Happy Horse (up to 9 images, 95 at 720p).
Sources
- fal, HappyHorse overview (model description, seven lip-sync languages, $0.14 and $0.28 per second): https://fal.ai/happyhorse, checked 2026-10-18
- fal, Happy Horse reference-to-video endpoint (1 to 9 images, 400 px shortest side, 10 MB, character1 to character9, 3 to 15 s, 720p and 1080p, five aspect ratios, prices): https://fal.ai/models/alibaba/happy-horse/reference-to-video, checked 2026-10-18
- fal, Happy Horse video-edit endpoint (up to 5 reference images, input 3 to 60 s, 100 MB, MP4 and MOV, two price statements): https://fal.ai/models/alibaba/happy-horse/video-edit, checked 2026-10-18
- Alibaba Cloud, "Alibaba Rolls Out HappyHorse 1.0 in Limited Beta", 2026-04-28: https://www.alibabacloud.com/blog/603068, checked 2026-10-18
- Alibaba Cloud, HappyHorse 1.1 upgrade post, 2026-06-23: https://www.alibabacloud.com/blog/happyhorse-gets-stronger-motion-expressiveness-higher-generation-consistency-and-enhanced-visual-quality_603293, checked 2026-10-18
- Alibaba Cloud Model Studio, HappyHorse reference-to-video API reference (happyhorse-1.1-r2v, 1 to 9 images, 480P/720P/1080P, nine aspect ratios): https://www.alibabacloud.com/help/en/model-studio/happyhorse-reference-to-video-api-reference, checked 2026-10-18
- Kling, VIDEO 3.0 Model User Guide (3 to 15 s, 1080p and 720p, native audio): https://kling.ai/quickstart/klingai-video-3-model-user-guide, checked 2026-10-18
- Kling, VIDEO O1 User Guide (1 to 7 reference images or elements, up to 4 images per element, 3 to 10 s, 1080p and 720p): https://kling.ai/quickstart/klingai-video-o1-user-guide, checked 2026-10-18
- fal, Kling v3 Standard image-to-video endpoint ($0.084 silent and $0.126 with audio per second, start image required): https://fal.ai/models/fal-ai/kling-video/v3/standard/image-to-video, checked 2026-10-18
- fal, Kling O1 reference-to-video endpoint ($0.112 per second, 7 inputs total, 5 or 10 s): https://fal.ai/models/fal-ai/kling-video/o1/reference-to-video, checked 2026-10-18
- 8frame prices, defaults and limits: the 8frame app, checked 2026-10-18
- Audio-track share: 8frame production database, completed generations in the 120 days to 2026-10-02, template examples excluded
Board both, attach the same references, keep the one that works. Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.