Jimeng AI (ByteDance) AI Comic Drama Platform Research

1. Introduction

Jimeng AI is an AIGC creation platform launched by ByteDance's CapCut (Jianying) team, formerly named "CapCut Dreamina" and officially renamed "Jimeng" in May 2024. In the AI comic drama space, Jimeng's distinctive position is that it is the only platform that simultaneously holds models, creation tools, and distribution channels: on the model side, it integrates the Seedance series of video models from ByteDance's Seed team; on the tool side, it offers Smart Canvas and the "AI Studio"; on the channel side, it connects directly to Douyin and Hongguo short dramas.

1.1. Developer and Version History

1.2. Positioning

Jimeng's product positioning is "a one-stop AI content creation platform for the general public and professional creators." In the AI comic drama pipeline, it covers the full span from script imagery, character design, and storyboard frames to final video generation, but does not cover distribution or revenue sharing (handled by Douyin and Hongguo).

Compared with other platforms in the same group, Jimeng's positioning has two notable characteristics:

• Built-in channel: Douyin includes built-in AI gameplay templates, AI avatars, and duet features, with Xiao Yun Que AI Short Drama Agent as a separate short-drama production entry point. Generated content can flow directly into the Douyin distribution system, an advantage that startup platforms cannot replicate.
• Consumer-friendliness over programmability: Jimeng prioritizes web and mobile interaction, with developer capabilities spilling out through the Volcano Engine Ark API rather than being offered as CLI / workflow orchestration within Jimeng itself. This makes it better suited for "human operation" than "Agent operation".

1.3. Pricing

Note: the figures above are quoted from a CITIC Securities research report and could not be directly verified on Jimeng's official website, so they are [To be verified]. Another secondary source claims Jimeng Premium membership at 499 RMB/month equates to roughly 0.41 RMB/second, and reports "successive price increases in April, with 15-second video cost rising from 5 RMB to 40 RMB (C-end)", also. Platform pricing in this group was adjusted multiple times during 2026, so citations should note "as of September 2026".

1.4. Open Forms

2. Glossary

3. Feature Description

3.1. Video Generation

After integrating Seedance 2.5 on August 5, 2026, Jimeng's video generation capability has further aligned with ByteDance's Seed team's latest models in terms of duration, resolution, and consistency, but the specific magnitude of parameter improvements has not been publicly quantified, marked [To be filled].

3.2. Image and Smart Canvas

Smart Canvas is the most engineering-valuable module in Jimeng's AI comic drama pipeline; it internalizes "post-generation image retouching" from an external process into an in-platform capability:

Text-to-image and image-to-image models were upgraded to 3.0/3.1 in September 2025, and image model 2.1 was released in December 2024.

3.3. Narrative and Collaboration Tools

• AI Studio: the upgraded form of the web client since December 2025, which, together with the "Dimension Fold" micro-drama initiative, organizes scattered one-off generations into project-level workspaces.
• Smart Multi-Frame 2.0: launched at the same time as AI Studio, strengthening multi-image conditional input.
• Little Octopus Octo (April 9, 2026): Jimeng's first collaborative AI narrative creation tool, elevating "narrative" from a one-off generation into a collaborable object.
• Xiao Yun Que AI Short Drama Agent: an independent Agent entry point for short dramas, handling the automation from script to storyboard.
• OmniHuman / Motion imitation: released the OmniHuman multimodal video model in February 2025, strengthened motion imitation in March, and upgraded to 1.5 in September, used for character performance and talking-head content.

4. Platform Architecture

图 4-1｜即梦 AI 四层平台架构：从分发与生态到治理

数据来源：基于本文分析绘制的示意图。

4.1. Overall Architecture

┌─────────────────────────────────────────────────────────────┐
│  分发与生态层  抖音 / 红果 / 小云雀 / 火山方舟 API / 大客户方案  │
├─────────────────────────────────────────────────────────────┤
│  工具与工作区层  AI 片场 · 智能画布 · 小章鱼 Octo · 智能多帧    │
├─────────────────────────────────────────────────────────────┤
│  模型层  Seedance 2.5 / 2.0 · Seedream 5.0 Lite · OmniHuman  │
│          视频 3.5 Pro · 图片 3.0/3.1 · Seaweed                │
├─────────────────────────────────────────────────────────────┤
│  治理层  数字人分身认证 · 真人素材限制 · AI 生成合成内容标识      │
└─────────────────────────────────────────────────────────────┘

Jimeng's architecture is a typical "big-tech vertical integration": the model layer is jointly supplied by ByteDance's Seed team and the CapCut team, the tool layer is built in-house by the CapCut team, and the distribution layer is handled by ByteDance's content ecosystem. The advantage of this architecture is a short pipeline and fast iteration; the cost is that each layer's capability boundaries are determined by ByteDance's internal product divisions, and external users cannot replace any single layer.

4.2. Model Layer

4.3. Tool and Workspace Layer

The tool and workspace layer is where Jimeng's incremental value lies relative to pure model services (such as the Seedance API), and it is its primary carrier at L2/L3. Core components are: AI Studio (project workspace), Smart Canvas (image editing), Smart Multi-Frame (multi-image condition), and Little Octopus Octo (narrative collaboration).

4.4. Distribution and Ecosystem Layer

• Built into Douyin: AI gameplay templates, AI avatars, and duet features.
• Xiao Yun Que AI Short Drama Agent: independent entry point for short-drama production.
• Film & TV institution cooperation: Huace Film & TV and Linmon Pictures announced cooperation to explore short-drama creation.
• Co-production: launched the AIGC sci-fi short-drama series 《Sanxingdui: The Future Revelation》 with Douyin and Bona Film Group.
• Industry activities: cooperated with the Busan International Film Festival; co-launched the InnoAsia AI Film International Summit with Volcano Engine and Shanghai Film (Group); cooperated with the ADC Annual Design Awards to launch a special award for AI visual design.
• Large-customer solutions: Bona Film Group, BlueFocus, and others enjoy dedicated inference compute and customized knowledge bases.

5. Harness Design

5.1. L1 Context Engineering Layer

Jimeng's context engineering capability is embodied in the capacity and organization of four-modality mixed input. Seedance 2.0 supports simultaneous input of four modalities — image, video, audio, and text — with up to 9 images, 3 videos, 3 audio tracks, plus a text prompt per generation. For AI comic drama, this means a single generation can simultaneously anchor: the character reference image (who), the scene reference image (where), the camera-movement reference video (how to shoot), the voice reference audio (what sound), and the text storyboard (what to do).

The accompanying image model Seedream 5.0 Lite has real-time web retrieval and intelligent logical reasoning capabilities, expanding the source of context from "user uploads" to "real-time retrieval".

Gap: Jimeng has not disclosed mechanisms for context compression, prioritization, or cache reuse. Compared with explicit compression schemes such as Hailuo AI's H3-Context-IR (compressing roughly 100,000 tokens to about 4,000 tokens), Jimeng follows the "expand input capacity" route rather than the "compress context" route. When reference material exceeds the 9 images + 3 videos + 3 audio clips limit, users must trim it themselves, with no public tiering strategy.

5.2. L2 Tool and Execution Layer

Jimeng's L2 is characterized as "rich tools but closed". Smart Canvas covers the most frequent image-retouching needs in comic drama production, a layer that startup platforms generally lack; however, these tools are exposed through GUI interaction with no corresponding programmatic interface, so they cannot be directly orchestrated by external Agents. Developers who want to programmatically invoke them can only detour through the Volcano Ark API, which covers only the model layer and not Smart Canvas capabilities.

5.3. L3 Orchestration and Control Layer

Jimeng's investment in L3 is understated but clear:

• Multi-shot storytelling: Seedance 2.0 comes with built-in planning and can complete shot transitions from wide to medium to close-up within a single generation. This embeds orchestration capability into the model — users need not explicitly orchestrate, but also cannot explicitly intervene.
• Little Octopus Octo: the collaborative AI narrative creation tool launched in April 2026, Jimeng's first explicit productization attempt at "narrative-level orchestration".
• Xiao Yun Que AI Short Drama Agent: turns the script-to-storyboard pipeline into an Agent-driven process.

Assessment: Jimeng's orchestration relies primarily on "implicit in-model planning + upper-level Agent encapsulation", lacking an explicit, editable intermediate representation (such as an exportable storyboard table or an editable node graph). This contrasts with PixVerse Canvas and ComfyUI node graphs. For users, the benefit is a fast learning curve; the cost is difficulty in managing fine-grained versioning and regression.

5.4. L4 Memory and State Layer

This is Jimeng's relatively weakest layer, and a direct footnote to this group's core thesis.

The L4-related capabilities Jimeng publicly discloses include:

• Stable preservation of character features (a model-side capability, mentioned in the Volcano Engine Seedance 2.0 model description).
• Asset library / material library (AI Studio, as a project-level workspace, can host material organization).
• Smart Canvas's multi-layer editing provides de facto layer-level state.

But there are three explicit gaps:

1. Portfolio scope not disclosed. Compared with Vidu's "subject library (characters/props/scenes, up to 7 reference images)", Jimeng does not disclose the capacity, persistence scope, or cross-project reuse rules of its asset library; this group marks it as [To be filled].
2. No shot state machine. Cross-shot plot state (character injuries, character relationships, timeline progression) has no public persistence mechanism.
3. Cross-session continuation depends on manual work. AI Studio is a project workspace, not a state machine; a 100-episode project on Jimeng still requires users to maintain the plot-state table themselves.

For AI comic drama, the scenario that puts the most pressure on L4, Jimeng provides a strong "channel and tools" answer but leaves the "state management" problem to the user.

5.5. L5 Evaluation and Observation Layer

Jimeng's platform side shows no public evaluation and observation products (such as trajectory tracking, regression sets, or availability statistics dashboards); this group marks it as [To be filled]. All currently available evaluation signals come from outside the platform:

This shows that Jimeng's L5 is "evaluation borne by the market rather than the platform": whether a work is good depends not on platform metrics but on Douyin play counts and revenue-sharing amounts. For high-frequency continuous production, this is an obvious engineering gap.

5.6. L6 Governance and Security Layer

Jimeng is the most distinctive governance sample in this group — it has both the strictest real-person appearance controls and a record of the most serious compliance incident.

Positive mechanisms:

• Digital avatar verification: since February 2026, it has restricted the use of real-person material and introduced a digital avatar verification mechanism. In the Jimeng App and Doubao App, if users want their real-person appearance on screen, they must record their own appearance and voice to complete real-person verification before creating a digital avatar.
• Real-person faces disabled: when using Seedance 2.0 on Jimeng's Web client, Xiao Yun Que, and other platforms, the system explicitly indicates that real-person faces are not yet supported as reference material.
• Regulatory constraints: subject to the 《Interim Measures for the Management of Generative AI Services》(effective August 2023), it must complete algorithm filing, uphold core values, and label synthetic media.

Negative event:

On April 28, 2026, Jimeng AI's website was lawfully penalized by cyberspace authorities for failing to effectively implement the labeling requirements for AI-generated synthetic content.

This incident is key evidence for understanding the importance of L6 in the AI comic drama scenario: even a leading big-tech platform with a full legal and compliance team can see its labeling pipeline break. It proves that enforcement of the 《Measures for the Labeling of AI-Generated Synthetic Content》 and GB 45438—2025 is not a one-time "configure at launch" task but a pipeline capability that needs continuous verification.

Under GB 45438—2025, the explicit label on video content must contain an "artificial intelligence (or AI)" element plus a "generated/synthetic" element, positioned on the edge or corner of the starting frame (may also include the end or middle), with a text height no less than 5% of the shortest side of the frame, and a duration of no less than 2 seconds at normal playback speed; the metadata implicit label must be written into an AIGC structure containing fields such as Label / ContentProducer / ProduceID / ContentPropagator / PropagateID.

5.7. Six-Layer Capability Matrix

6. Use Cases

6.1. Case 1: AIGC Sci-Fi Short-Drama Series 《Sanxingdui: The Future Revelation》

• Background: film-grade AIGC content must balance visual spectacle with narrative coherence, a typical scenario for testing a platform's continuous production capability.
• Solution: Jimeng, together with Douyin and Bona Film Group, launched the AIGC sci-fi short-drama series 《Sanxingdui: The Future Revelation》.
• Result: the work successfully entered the film-grade AIGC tier; specific production duration, cost, and playback data have not been publicly disclosed, marked [To be filled].

6.2. Case 2: Visual Effects Generation for the 2026 CCTV Spring Festival Gala

• Background: Gala-level content imposes extremely high demands on image quality, stability, and delivery deadlines, and cannot be produced through a "trial and error" approach.
• Solution: Jimeng cooperated with CCTV to generate visual effects for the 2026 CCTV Spring Festival Gala programs.
• Result: it validated Jimeng's usability in high-spec delivery scenarios; the specific number of generated shots and production duration have not been publicly disclosed, marked [To be filled].

6.3. Case 3: Dedicated Inference Solutions for Institutional Clients

• Background: the demands of large film and marketing institutions differ from those of C-end users — they need stable compute guarantees and private knowledge/material accumulation.
• Solution: major clients such as Bona Film Group and BlueFocus enjoy dedicated inference compute and customized knowledge bases.
• Result: this is one of Jimeng's rare enterprise-grade deliveries at L4 (customized knowledge base) and L6 (dedicated resource isolation); specific SLAs and isolation mechanisms have not been publicly disclosed, marked [To be filled].

6.4. Case 4: Compliance Incident and Rectification

• Background: the 《Measures for the Labeling of AI-Generated Synthetic Content》 took effect on September 1, 2025, and the supporting mandatory national standard GB 45438—2025 imposes quantitative labeling requirements.
• Solution: on April 28, 2026, Jimeng AI's website was lawfully penalized by cyberspace authorities for failing to effectively implement the labeling requirements for AI-generated synthetic content, and was subsequently rectified.
• Result: this incident became this group's only public penalty case, directly influencing the evaluation baseline for each platform's L6 maturity — Jimeng appears in the same matrix in the dual capacity of "strongest governance sample" and "only penalized platform".

7. Summary

7.1. Strengths

1. Unique channel closed-loop: the model + creation tools + Douyin/Hongguo distribution are one integrated whole, a structural advantage that startup platforms cannot replicate.
2. Tool layer best fits comic drama retouching needs: Smart Canvas's multi-image fusion, localized repainting, expansion, removal, background removal, and multi-layer editing map precisely to the most frequent manual-repair scenarios in AI comic drama.
3. Fastest model iteration cadence: from Seaweed in 2024-11 to Seedance 2.5 in 2026-08, the video model iterated at least four generations in two years.
4. Strictest real-person appearance controls: digital avatar verification + real-person faces disabled is the earliest and clearest real-person appearance governance in the industry.

7.2. Limitations

1. L4 is an obvious shortcoming: no public asset library capacity or shot state machine; the state management for continuous production relies entirely on users building it themselves.
2. Tools are not programmable: core tools such as Smart Canvas have no programmatic interface and cannot be integrated into one's own Agent or CI.
3. L5 is absent: no public evaluation and observation products; quality judgment is left entirely to the market.
4. Opaque pricing: membership pricing has two sets of secondary figures and was adjusted multiple times during 2026, making cost accounting difficult.
5. Has a compliance-incident record: penalized in April 2026, indicating that labeling-pipeline reliability requires additional review by users.

7.3. Applicability Boundary

7.4. Selection Recommendations

• The core reason to choose Jimeng is channel and tools, not Harness completeness. If your main distribution front is Douyin/Hongguo, Jimeng's distribution advantage outweighs its L4 shortcoming.
• You must build a project-level state layer on top of Jimeng: storyboard tables (structured JSON), character cards (with reference images and text anchors), scene cards, and a plot state machine. Do not expect AI Studio to manage plot state for you.
• You must build your own compliance review: after export, verify the explicit label (starting frame, corner position, text height no less than 5% of the shortest side, duration no less than 2 seconds) and the AIGC metadata implicit-label fields; do not assume the platform has fully implemented them.
• Do cost accounting with the most conservative figures: use the 499 RMB/month Premium membership tier and reserve more than 30% rework redundancy; do not budget using the industry's optimistic "1,000 RMB per work" estimate.

Information Gap Statement

1. Real-time figures on Jimeng AI's official pricing page: the two sets of figures — 69/199/499 RMB/month and 79 RMB/month — are both quoted from secondary research reports and could not be directly verified on Jimeng's official website.
2. The specific form of Jimeng's asset library: capacity limit, persistence scope, cross-project reuse rules, and whether cross-episode references are supported all have no public documentation found.
3. Parameter changes after Seedance 2.5 was integrated into Jimeng: the specific improvement magnitude for duration, resolution, and consistency metrics has not been officially quantified.
4. Jimeng's platform-side evaluation and observation capabilities: whether mechanisms such as availability statistics, generation-history tracking, and regression sets exist has no public result.
5. Whether Jimeng has built-in GB 45438—2025-compliant metadata implicit labels: only the negative evidence of being penalized exists, with no positive compliance statement found.
6. Technical details of institutional-customer solutions: the isolation method of dedicated inference compute, the capacity and retrieval mechanism of customized knowledge bases, and SLA terms all have no public result.
7. The functional boundary of Little Octopus Octo: as a "collaborative AI narrative creation tool", its collaboration model, export format, and whether external integration is supported have not been described in detail.

8. References

1. Baidu Baike 《Jimeng AI》 — https://baike.baidu.com/item/%E5%8D%B3%E6%A2%A6/64611350
2. Baidu Baike 《Seedance 2.0》 — https://baike.baidu.com/item/Seedance%202/67291551
3. Baidu Baike 《AI Comic Drama》 — https://baike.baidu.com/item/AI%E6%BC%AB%E5%89%A7/68788906
4. Baidu Baike 《Keyframe Animation》 — https://baike.baidu.com/item/%E5%85%B3%E9%94%AE%E5%B8%A7%E5%8A%A8%E7%94%BB/10223838
5. Jimeng AI official website — https://jimeng.jianying.com/
6. Volcano Engine Ark console, Doubao-Seedance-2.0 series — https://console.volcengine.com/ark/region:ark+cn-beijing/model/detail?Id=doubao-seedance-2-0
7. Four departments including the Cyberspace Administration of China, 《Measures for the Labeling of AI-Generated Synthetic Content》(National Cyberspace Administration Notice No. 2 of 2025) — https://www.cac.gov.cn/2025-03/14/c_1743654684782215.htm
8. Mandatory national standard GB 45438—2025 《Network Security Technology — Methods for Labeling AI-Generated Synthetic Content》 — https://www.tc260.org.cn/upload/2025-03-15/1742009439794081593.pdf
9. CCTV News, 《How will labeling, dissemination, and review of AI-generated synthetic content develop?》 — https://news.cctv.cn/2025/03/15/ARTI3wMX1ohsE7LsLT4qBVrv250315.shtml
10. 21st Century Business Herald, 《1 RMB per second! ByteDance's Seedance 2.0 pricing released》 — https://m.21jingji.com/article/20260305/herald/9065ec418e98b284842d12cab6aa23fb.html
11. Toutiao, 《The monetization logic of AI comic dramas can be summed up as "one foundation, three wings"》 — https://www.toutiao.com/article/7678962433607664163
12. The Paper, 《Producing 1 episode for 1,914 RMB? Comic dramas remain trapped in hidden costs》 — https://www.thepaper.cn/newsDetail_forward_33993493
13. Tencent News, 《Douyin tops the charts, Hongguo is fierce: ByteDance's traffic business and its next exam》 — https://new.qq.com/rain/a/20260909A0EWUR00
14. KOCPC (English) retelling of the DataEye 2026 H1 AI short-drama report — https://en.kocpc.com.tw/archives/25203
15. EPWK, 《AI Comic Drama Storyboard Design Guide》 — https://gonglue.epwk.com/322844.html
16. AI Wiki 《Doubao》 — https://aiwiki.ai/wiki/doubao