Jimeng (Dreamina)
Last reviewed
Sources
19 citations
Review status
Source-backed
Revision
v2 · 2,007 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
19 citations
Review status
Source-backed
Revision
v2 · 2,007 words
Add missing citations, update stale details, or suggest a clearer explanation.
Jimeng (Chinese: 即梦, "instant dream"), branded internationally as Dreamina, is a consumer AI image and video generation platform developed by ByteDance that turns text prompts, reference images, or both into still images and short video clips. [1][2] It is part of ByteDance's Jianying creative product family, the video editor marketed outside China as CapCut, is operated by ByteDance subsidiary Shenzhen Faceu Technology, and is powered by ByteDance's own foundation models served through the company's Volcano Engine (Volcengine) cloud: the Seedream line for images and, for video, the Doubao video models (PixelDance and Seaweed) followed by the Seedance line. [3][4][9]
In China the product carries the 即梦 / Jimeng brand and is reachable at jimeng.jianying.com and through a standalone mobile app; internationally the same product is offered under the Dreamina name within the CapCut suite. The desktop version launched in May 2024 and the standalone Android app on 31 July 2024, with an iOS release for China following in early August 2024. [9] By 2026 ByteDance was using both names together, often writing the video model as "Dreamina Seedance," and Jimeng / Dreamina had become one of the most-used AI creation tools competing with Kuaishou's Kling and MiniMax's Hailuo. [4][5]
Jimeng is positioned as a one-stop creation tool rather than a single generator: it combines text-to-image, image editing, text-to-video, and image-to-video in one place, along with an editing canvas and digital-avatar features. [1][6] The stated audience is short-video and image creators, especially those who also publish on ByteDance's Douyin (the Chinese counterpart of TikTok), and the tool is meant to shorten the path from an idea to a finished social-media clip. [7] ByteDance built a deliberate content pipeline around the product in which Dreamina generates, CapCut edits, and Douyin or TikTok distributes. [7]
ByteDance has described the broader strategy behind the product as a bet that visually rich, low-effort creation tools, rather than text chatbots, are the more natural mass-market form for generative AI. The effort was started by Kelly Zhang, a former chief executive of Douyin, after she moved to the CapCut team in February 2024, and ByteDance has internally framed Dreamina as a candidate to become the "TikTok of the AI era." [7]
The naming around this product is layered, and several sources use the names loosely, so the verifiable history is worth stating carefully. The platform began under the CapCut creative brand. In late March 2024 a tool then called Jianying Dreamina (剪映Dreamina), or CapCut Dreamina in English, entered open beta. On 9 May 2024 the Chinese version adopted the brand name 即梦 (Jimeng) and fully launched its AI image and AI video features. [2][8] The international version offered through CapCut kept the Dreamina name. As a result, Jimeng and Dreamina refer to the same line of products from the same team, with Jimeng being the China-facing brand and Dreamina the overseas brand; primary press has stated plainly that Dreamina is "known in China as Jimeng AI." [7]
ByteDance also released Jimeng as a separate mobile app. The standalone Android version appeared on 31 July 2024, followed shortly by an iOS release for users in China, which is the launch many Western outlets reported in early August 2024 as ByteDance's answer to OpenAI's then-unreleased Sora. [9][10] Because the China brand (Jimeng), the overseas brand (Dreamina), and the underlying models (Seedream, Seedance) each have their own names, coverage often blends them; the safest reading is that "Jimeng" and "Dreamina" name the product, while "Seedream" and "Seedance" name the models inside it.
Jimeng / Dreamina bundles generation and editing tools that share the same underlying models. The capabilities below are drawn from the official Jimeng site and product coverage.
| Feature | What it does |
|---|---|
| Text-to-image | Generates still images from a text prompt, with style presets and templates |
| Image-to-image and editing | Style transfer, background replacement, and subject-preserving edits |
| Smart canvas | A multi-layer workspace with local inpainting (partial repaint), one-click image expansion (outpainting), object removal, and cutout |
| Text-to-video | Produces a short clip from a text description |
| Image-to-video | Animates a still image into a clip |
| First-frame and last-frame control | Lets the user fix the opening and closing frames of a generated video |
| Camera and motion control | Cinematic camera moves and natural motion within a shot |
| Digital human / avatar | Creates a lifelike avatar that can be driven with lip-synced speech and body movement |
| Story and storyboard creation | Tools to plan shots and assemble multi-shot sequences |
The image side emphasizes Chinese-language prompt understanding and accurate rendering of Chinese text inside images, a capability ByteDance has highlighted as a differentiator. [7] The video side covers both generation from scratch and animation of uploaded images, with controls for camera movement, and the digital-human feature combines a generated or uploaded portrait with the video pipeline to produce talking-avatar clips with synchronized lips and gestures. [6][1] The smart canvas groups the common image-editing operations (repaint, expand, remove, cut out) into one editor so users do not switch between separate tools. [11]
Jimeng / Dreamina does not train its own models in isolation; it surfaces ByteDance's foundation models, which are developed by the company's Seed research group and served through Volcano Engine. The model lineage has moved quickly.
For video, ByteDance's first generation came from the Doubao video models. On 24 September 2024 Volcano Engine unveiled two Doubao video generation models, PixelDance and Seaweed, for invited enterprise testing, marking the company's formal entry into the AI video race; ByteDance said the same DiT (diffusion transformer) work was optimized for products including Jimeng AI. [3][12] PixelDance was aimed at complex, sequential motion and multi-subject interaction, while Seaweed was the model brought to Jimeng / Dreamina's own users: on 8 November 2024 the platform opened the standard edition of the Doubao Seaweed model to users, generating a five-second clip in about a minute. [13] ByteDance later introduced the Seedance line as its main video series, including Seedance 1.0, the intermediate Seedance 1.5 Pro, and Seedance 2.0, a multimodal model released on 12 February 2026 that generates synchronized audio and video together and is exposed to consumers as "Dreamina Seedance 2.0." [4][14][17]
For images, the platform is powered by the Seedream series. By early 2026 the current image model was Seedream 5.0, which ByteDance rolled out inside its own apps (Jianying, CapCut, and Jimeng AI) in February 2026, adding multi-step visual reasoning and an online-search capability on top of the earlier Seedream releases. [15][16]
The table summarizes the model components and where they sit in the stack.
| Component | Type | Role in Jimeng / Dreamina |
|---|---|---|
| Seedream (to 5.0) | Image generation | Powers text-to-image and image editing |
| Doubao PixelDance | Video generation | Early Doubao video model; complex and multi-subject motion |
| Doubao Seaweed | Video generation | First video model opened to Jimeng / Dreamina users (Nov 2024) |
| Seedance (to 2.0) | Video generation | Current video series; text-to-video and image-to-video, audio-video in 2.0 |
| Doubao brand | Model branding | Developer-facing naming (e.g. Doubao-Seedance) on Volcano Engine |
| Volcano Engine | Serving platform | Hosts and serves the models; provides the developer API |
The models are also built by ByteDance Seed and offered to developers through Volcano Engine in China and BytePlus internationally, where the same systems appear under Doubao-prefixed identifiers. [4][16]
Dreamina Seedance 2.0 is the version of ByteDance's video model that the company began rolling out inside CapCut on 26 March 2026, the first time the Seedance 2.0 system reached CapCut's global creators after launching as a model on 12 February 2026. [5][17] ByteDance described it as letting creators "draft, edit, and sync video and audio content by using prompts, images, or reference videos," and said the model "works without reference images, even if the creator only uses a few words to describe the scene." [5] At launch it supported clips of up to 15 seconds across six aspect ratios. [5]
The CapCut rollout was phased and began with seven markets: Brazil, Indonesia, Malaysia, Mexico, the Philippines, Thailand, and Vietnam, with the company adding the United States, Japan, Europe, and other regions over the following weeks. [5][18] ByteDance attached safety restrictions to the launch: the model will not generate videos from images or clips that contain real faces, it blocks unauthorized use of intellectual property, and every output carries an invisible watermark so content made with the model can be identified when shared off-platform. [5][18]
Jimeng arrived during an intense burst of AI video activity among Chinese technology companies. At its August 2024 app launch, Western reviewers judged its text-to-video output as capable but still showing common artifacts such as jitter and stilted movement, and not yet at Sora's level. [9][10] Its closest domestic rivals were Kuaishou's Kling, which had launched earlier in 2024, along with Zhipu AI's Ying and Shengshu's Vidu; MiniMax's Hailuo is the other major Chinese consumer video generator it is regularly compared against. [9][7]
Pricing at the Chinese launch was a monthly subscription of 69 yuan (about 9.70 US dollars), with a free tier allowing roughly 80 images or 26 videos and paid plans supporting around 2,000 images or nearly 170 videos per month. [9][19] The platform's scale is best understood through its ties to ByteDance's larger products: CapCut reached around 170 million monthly active users by October 2024, second worldwide behind ChatGPT at the time, and ByteDance has treated Dreamina / Jimeng as a strategic creative surface within that ecosystem rather than a standalone bet. [7] Specific standalone user figures for Jimeng / Dreamina are not reliably published, so they are omitted here.
As the underlying models improved, Jimeng / Dreamina's competitive standing rose with them. When Seedance 2.0 reached Dreamina and CapCut in early 2026, independent human-preference leaderboards placed the "Dreamina Seedance 2.0" entry at or near the top among video models that also generate audio, and the international rollout through CapCut's paid tiers expanded the product into Southeast Asia, Latin America, and other markets, even as OpenAI was winding down its Sora app in the same period. [14][16]