AI Comic Drama Platform Market Research (Group Overview)

1. Introduction

AI comic drama is one of the scenes in the AIGC content industry where engineering maturity is rising fastest and that first exposed the contradiction of "model capability ≠ production readiness." It is neither static comics nor live-action short drama, but an intermediate form with comic storyboards as its skeleton and dynamic audiovisual language as its flesh. In the first half of 2026, domestic platforms cumulatively launched 367,000 micro short dramas, of which AI content accounted for over 74% (China Netcasting Services Association data, as cited by KOCPC); AI comic drama has moved from "tech demo" to "industrial assembly line."

But the flip side of the assembly line is extremely low commercial conversion: DataEye's 2026 H1 AI Drama/Comic Drama Data Report shows that among 221,900 AI short dramas, only 1,055 exceeded 100 million plays, a hit rate of 0.47%; measured against a break-even line of 50 million plays, the break-even rate is below 1.3%, roughly only 1 in 77 titles recoups its cost. Oversupply and a scarcity of hits coexist, showing that the decisive factor in industry competition has shifted from "can we generate a good-looking video" to "can we stably, continuously, and controllably produce a hundred episodes."

This is precisely where AI Harness (the artificial intelligence harness layer) comes in. Per the unified definition in this project's parameter card, Harness sits between the foundation model and real business systems, converting the model's uncertainty into engineering predictability. In the AI comic drama scene, "uncertainty" concretely manifests as: the same character looking different in episode 1 versus episode 47, the same scene flickering between two shots, and plot state being lost when continuing across sessions. None of these problems can be solved by switching to a stronger model; they can only be solved by Harness's context engineering, tool orchestration, state management, evaluation regression, and governance mechanisms.

1.1. Research Scope and Subjects

This group selected 8 representative platforms/workflows, covering four organizational forms:

Selection basis: these 8 subjects cover all pronounced tier differences across the six capability layers L1 to L6 in this group's horizontal comparison matrix (see Section 4.2), and each has enough publicly verifiable information to stand on its own as a separate article.

1.2. This Group's Position in the AI Harness System

AI comic drama is a high-pressure sample for testing the Harness six-layer model. Compared with scenes such as software engineering and data analysis, it has four unique sources of pressure:

1. Extremely large state span: a 100-episode comic drama requires state to span hundreds to thousands of independent inference calls, real time spanning days to weeks, and multiple sessions and multiple collaborators.
2. No incremental correction: code can be patched and documents can have paragraphs revised, but once a video is rendered, changing a character's hairstyle means re-rendering the entire shot or even the whole scene.
3. Evaluation lacks an objective yardstick: code has tests and SQL has result sets, but "is this performance any good" and "does this character look alike" lack automated judgment means; evaluation relies heavily on manual review and external platform leaderboards.
4. Compliance is a precondition, not a post-hoc check: the Measures for Labeling AI-Generated Synthetic Content and the mandatory national standard GB 45438—2025 impose quantitative requirements on the explicit label position, font height, duration, and metadata fields of video; non-compliant content cannot be released or distributed.

Therefore, this group's documents do not evaluate platforms as "tools," but rather dissect each platform as an already-deployed Harness implementation: examining where it made real investment in each layer, where it left gaps, and who fills the gaps.

1.3. Dissection Method

The same dissection framework is applied uniformly to all 8 platforms to ensure horizontal comparability:

图 1-1｜Harness 六层模型：AI 漫剧场景的压力分布

数据来源：基于本文分析绘制的示意图。

2. Glossary of Terms

This table is the common terminology baseline for the whole group; the glossary sections in each platform document build on this baseline by adding platform-specific concepts.

3. Industry Status

3.1. Market Size and Production Capacity

Conclusion: production capacity is severely oversupplied and unit price is collapsing. The industry is moving from the "scrambling for capacity" phase into the "scrambling for certainty" phase—whoever can stably produce hits holds the bargaining power.

3.2. Cost Structure and Hidden Costs

It should be specifically pointed out: The Paper's RMB 1,000/title estimate has clear preconditions—the prompt is written extremely well, every image is usable on the first pass, and there is no need to repeatedly rework character consistency. The hidden costs in reality come precisely from here: character consistency is the biggest technical bottleneck, and solving it requires building character LoRAs, training character models, maintaining asset libraries, or manually retouching images one by one—costs that are systematically obscured by platform narratives.

CICC's estimate reveals the other side of the cost structure: the AI production stage concentrates 70%~80% of the industry's cost; virtual shooting lots cut scene-construction cost by up to 85%; digital human actor character-modeling cost is reduced by 80%; and VFX production cycles are shortened by up to 40%.

3.3. Revenue-Sharing Mechanism and Filing Threshold

Revenue-sharing formula: per-title comic drama free revenue share = incremental valid duration that month × per-duration unit price × comic drama type coefficient × copyright coefficient.

Revenue per 10,000 plays dropped from about RMB 100 to RMB 5~10. At the same time, platform-side minimum-guarantee policies are diverging as they contract:

• In 2026, Douyin Group announced a full-year minimum-guarantee budget of over RMB 1.5 billion for live-action short drama, with per-title guarantee amounts up about 60% year-over-year, script guarantee fees of RMB 20,000 to 100,000, and a revenue-share ratio of 15%~20%.
• In early 2026, Douyin Group's Short Drama Copyright Center launched a new incentive mechanism for AI-likeness character drama: tiered guarantees of RMB 20,000 to 80,000 plus a uniform 20% permanent revenue share.
• In May 2026, Hongguo clarified it had not "completely abolished" guarantees; only some premium scripts retain them.
• During the 2026 Spring Festival season, Hongguo's platform-wide total views reached 33.61 billion, with 1.28 billion total viewers; "Shengxia Fendela" received a revenue share of over RMB 16 million.
• In July 2026, AI comic dramas occupied four of the top five spots on Hongguo's trending list.

On the regulatory side: in January 2026 the filing threshold for key micro short dramas was raised from RMB 1 million to RMB 3 million; that month 132 key titles were filed, down nearly half month-over-month. In December 2024, the NRTA's Internet Audiovisual Division and the Beijing Municipal Radio and Television Administration jointly summoned Hongguo's leaders; Hongguo suspended new releases for 5 days, took down 279 non-compliant micro short dramas, and changed 1,080 non-compliant titles.

3.4. Compliance Baseline

All platform documents in this group base their L6 dissection on the following two documents as the unified baseline.

The Measures for Labeling AI-Generated Synthetic Content (Guoxintongzi [2025] No. 2), issued on March 7, 2025 and effective September 1, 2025:

• Article 4: Video must have a prominent explicit label added at an appropriate position on the opening frame and around the playback area; it may also be added at the end and in the middle.
• Article 5: An implicit label must be added to the file metadata, including attribute information, the service provider's name or code, and the content number.
• Article 6: Distribution platforms must verify the implicit metadata label; if it is not found but the user declares the content, it is marked as "may be" (AI-generated); if traces are detected, it is marked as "suspected to be" (AI-generated).
• Article 10: Users publishing generated synthetic content should proactively declare it and use the labeling feature; no organization or individual may maliciously delete, alter, forge, or conceal the label.

The mandatory national standard GB 45438—2025 Cybersecurity Technology—Methods for Labeling AI-Generated Synthetic Content:

• The explicit label on video content must contain the "artificial intelligence (or AI)" element plus the "generated/synthesized" element.
• The label is located on the video's opening frame (may include the end and middle), positioned at the edge or corner of the frame.
• Text height must not be less than 5% of the shortest side length of the frame; at normal playback speed its duration must be no less than 2 seconds.
• Metadata implicit label field format: {"AIGC": {"Label": "...", "ContentProducer": "...", "ProduceID": "...", "ReservedCode1": "...", "ContentPropagator": "...", "PropagateID": "..."}}; Label values: belongs = 1, may be = 2, suspected = 3.

A negative case: on April 28, 2026, the Jimeng AI website was lawfully investigated and punished by cyberspace authorities for failing to effectively implement the labeling requirements for AI-generated synthetic content. This is the only public penalty event targeting an AI comic drama-related platform this group found, and it is the direct basis for treating L6 as a "hard constraint" rather than a "bonus item."

4. Horizontal Platform Comparison Matrix

4.1. Basic Capability Comparison

4.2. Harness Six-Layer Maturity Comparison

4.3. Consistency and Governance Capability Comparison

5. Core Arguments of This Group

5.1. L4 Is the Layer under the Most Concentrated Pressure in the AI Comic Drama Scene

AI comic drama is the scene where L4 (memory and state) bears the greatest pressure among the Harness six layers. Cross-shot, cross-episode character appearance consistency, scene continuity, and plot-state continuity are, in essence, all cross-session state retention problems:

• The model itself is stateless. A video generation call does not remember what a previous call produced.
• Character consistency is not a "model capability" problem but a "state injection" problem: injecting the character anchor information (reference images, LoRA, style embeddings, text descriptions) into the context in the correct form and with the correct weight on every generation.
• Scene continuity and plot state are the same: what injuries the protagonist sustained at the end of the last episode, and what relationship they have with the antagonist—these must be explicitly persisted and retrieved for later generations.

Whoever makes L4 solid (character anchoring, asset libraries, shot state machines) can push AI comic drama from "can generate" to "can produce continuously." The horizontal comparison matrix confirms this: platforms that make L4 their core selling point (Vidu subject library, Hailuo @ reference system, Bairimeng's permanent character library, ComfyUI character image library + LoRA) have all carved out differentiated positions in their respective tracks; meanwhile, platforms where L4 remains mere marketing phrasing like "stable character feature retention" have continuous-production capability that depends heavily on users' manual remediation.

5.2. L6 Is a Hard Constraint That Cannot Be Bypassed

The Measures for Labeling AI-Generated Synthetic Content, effective September 1, 2025, together with the supporting mandatory national standard GB 45438—2025, turned labeling of AI-generated synthetic content from an "industry initiative" into a "precondition for release": the position, font height, and duration of explicit labels and the metadata fields of implicit labels all have quantified requirements; distribution platforms have a verification obligation; and users may not maliciously delete, alter, forge, or conceal labels.

This means that in the AI comic drama scene, Harness's L6 is not a "safety bonus item" but a first-class attribute of the output—as important as resolution and frame rate, and it must be completed automatically at the end of the generation pipeline, not left to manual post-production. The case of Jimeng AI being sanctioned on April 28, 2026 shows that even leading big-tech platforms can have their labeling pipeline break.

In addition, real-person likeness is a second red line: Doubao/Seedance explicitly prohibit using real-person faces as reference material, and real people appearing on screen must complete likeness and voice verification; Jimeng introduced a digital avatar certification mechanism starting February 2026.

5.3. L1 to L3 Determine the Ceiling of a Single Generation, L4 Determines the Ceiling of Continuous Production

This group's second observation: the industry's current competitive focus is shifting from L1~L3 to L4.

• L1~L3 are already highly homogenized: four-modal input (Seedance, Hailuo H3), smart storyboarding (Kling), multi-shot narrative (Seedance, Jimeng), and node-based workflows (PixVerse Canvas, ComfyUI) had become standard among leading platforms by 2026. These capabilities determine "how good a single video can be."
• L4 is not yet homogenized: the form, capacity, and cross-session persistence of asset libraries differ enormously—from "up to 7 reference images" (Vidu) to "relational context compressed from about 100k tokens to about 4k tokens" (Hailuo H3-Context-IR) to "workflow JSON as state" (ComfyUI); the implementation paths are entirely different. These capabilities determine "whether a hundred episodes can feel like a single drama."

Corresponding to selection: for single ad films, MVs, and concept films, strong L1~L3 alone suffices; for episodic continuous production, L4's weight should be significantly higher than L1~L3.

6. Selection Recommendations

General recommendations (independent of platform choice):

1. On top of any platform, build a "project-level state layer" of your own: storyboard tables (structured JSON), character cards (including reference images and text anchors), scene cards, and a plot state machine. Platform asset libraries are accelerators, not substitutes.
2. Turn compliance labeling into an automatic node at the end of the pipeline, not a manual checklist item. Explicit labels must be on the opening frame, at an edge/corner position, with font height no less than 5% of the shortest side, lasting no less than 2 seconds; implicit labels must be written into the AIGC metadata fields.
3. Build your own regression set: fix 20~50 representative shots (including close-ups, action, night scenes, and multiple characters in frame); after every platform version upgrade, rerun and compare the character consistency rate and availability rate. The industry currently lacks a unified benchmark, so building your own is the only viable path.
4. Include hidden costs in cost accounting: by industry reality, character-consistency rework and manual image retouching often exceed pure generation costs; do not budget based on the optimistic "RMB 1,000/title" estimate.

7. Document Navigation

8. Summary

The AI comic drama industry has already crossed the "can we generate" threshold and is diverging at the "can we produce continuously" threshold. The H1 2026 data gives a clear signal: production capacity piles up at a pace of one title every 36 seconds, with a hit rate of 0.47% and a break-even rate below 1.3%. In a market of oversupply and unit prices down more than 90%, the model of gambling on probability by generating and advertising more has failed, and certainty has become the only scarce resource.

After dissecting the 8 platforms through the Harness six-layer model, this group draws three conclusions:

1. L1 to L3 are already highly homogenized and are no longer a source of differentiation. Four-modal input, smart storyboarding, multi-shot narrative, and node-based workflows were already standard among leading platforms by 2026.
2. L4 (memory and state) is the layer under the most concentrated pressure and the least homogenized in the AI comic drama scene. The implementation paths of character anchoring, asset libraries, and shot state machines differ greatly across platforms and directly determine "whether a hundred episodes can feel like a single drama."
3. L6 (governance and safety) is a hard constraint, not a bonus item. The quantified labeling requirements of the Measures for Labeling AI-Generated Synthetic Content and GB 45438—2025, plus the case of Jimeng AI being sanctioned, show that the labeling pipeline must be built with the same life cycle as the generation pipeline.

The practical implication for engineering teams is: do not treat platforms as black-box tools; instead treat each platform as an already-deployed Harness, and know clearly which layer it does for you and which layer you must build yourself. Regardless of which platform is chosen, the project-level state layer (storyboard tables, character cards, scene cards, plot state machine) and the end-of-pipeline compliance labeling node should not be outsourced to the platform.

Information Gap Statement

1. Real-time figures on each platform's official pricing page: Jimeng AI (two sets of figures: RMB 69/199/499 per month and RMB 79 per month), Kling AI China site (RMB 19/58/198 per month), Hailuo AI (three sets: RMB 68/245/578/899/1399 per month, RMB 29/69 per month, and $9.99~$199.99), PixVerse (scraped by third party ToolChase on 2026-09-08), and Vidu (no full official price list) all could not be directly verified from their respective official sites, so all are marked [To be verified].
2. Official release date of Kling 3.0 and "up to 3 minutes": only second-hand sources exist (claiming a drama-specific edition on 2026-05-21); not confirmed on Kuaishou or Kling's official site.
3. Bairimeng AI's official pricing page, version number, and development team size: only aggregator and app-distribution site information exists, with low credibility; figures such as "RMB 200/minute," "50-minute long videos," and "95%+ cross-episode similarity" are of doubtful provenance.
4. Bairimeng AI's governance and safety mechanisms: no public search results at all; whether AI labeling is implemented, and whether permissions and audit capabilities exist, are all unknown.
5. ComfyUI's official version number: it uses rolling releases with no traditional semantic version number; this document substitutes GitHub star counts and node counts and does not state a version number.
6. Quantifiable third-party benchmarks for each platform's character consistency: apart from SuperClue (Vidu) and the Artificial Analysis ELO, there is no unified industry standard for a "cross-episode character consistency rate."
7. AGENTS.md / SKILL.md industry practices in the AI comic drama field: no public standard files were found; ComfyUI workflow JSON and PixVerse Skills are the closest forms.
8. Whether each platform natively embeds GB 45438—2025-compliant metadata implicit labels: apart from the negative evidence of Jimeng being sanctioned, no platform's positive compliance statement was found.
9. Market size and hit-rate data: mainly from DataEye reports relayed through second-hand sources (Baidu Baike, KOCPC, Keji Xiaomeng); the full text of DataEye's original report was not obtained for cross-validation.

9. References

1. Baidu Baike "AI Comic Drama" — https://baike.baidu.com/item/AI%E6%BC%AB%E5%89%A7/68788906
2. Baidu Baike "Jimeng AI" — https://baike.baidu.com/item/%E5%8D%B3%E6%A2%A6/64611350
3. Baidu Baike "Seedance 2.0" — https://baike.baidu.com/item/Seedance%202/67291551
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. 360 Encyclopedia "Keyframe" — https://baike.so.com/doc/6737995-32354145.html
6. The Measures for Labeling AI-Generated Synthetic Content, jointly issued by the Cyberspace Administration of China and three other departments (Guoxintongzi [2025] No. 2) — https://www.cac.gov.cn/2025-03/14/c_1743654684782215.htm
7. Chinese Government Website (Traditional Chinese version) "Measures for Labeling AI-Generated Synthetic Content" — https://big5.www.gov.cn/gate/big5/www.gov.cn/zhengce/zhengceku/202503/content_7014286.htm
8. Mandatory national standard GB 45438—2025 Cybersecurity Technology—Methods for Labeling AI-Generated Synthetic Content — https://www.tc260.org.cn/upload/2025-03-15/1742009439794081593.pdf
9. CCTV News "How will the labeling, dissemination, and review of AI-generated synthetic content develop?" — https://news.cctv.cn/2025/03/15/ARTI3wMX1ohsE7LsLT4qBVrv250315.shtml
10. People's Court of the 12th Division, Xinjiang Production and Construction Corps "New rules for labeling AI-generated content are here" — http://btd12szy.btcourt.gov.cn/article/detail/2025/08/id/8955655.shtml
11. The Paper "RMB 1,914 to produce one episode? Comic drama is still trapped in hidden costs" — https://www.thepaper.cn/newsDetail_forward_33993493
12. Tencent News "Hongguo short drama daily active users dwarf the combined total of iQiyi, Youku, Tencent Video, and Mango TV" — https://new.qq.com/rain/a/20260909A0FDHU00
13. Tencent News "Douyin tops the charts, Hongguo is aggressive: ByteDance's traffic business and the next challenge" — https://new.qq.com/rain/a/20260909A0EWUR00
14. Toutiao "The monetization logic of AI comic drama can be summarized as 'one foundation, three wings'" — https://www.toutiao.com/article/7678962433607664163
15. KOCPC (English) relay of the DataEye 2026 H1 AI short drama report — https://en.kocpc.com.tw/archives/25203
16. Keji Xiaomeng "The bubble has burst! AI comic drama exploded to 220,000 titles in half a year, 98.7% lost everything" — https://so.html5.qq.com/page/real/search_news?docid=70000021_6156a9ea29321852
17. 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
18. 21st Century Business Herald "RMB 1 per second! ByteDance Seedance 2.0 pricing released" — https://m.21jingji.com/article/20260305/herald/9065ec418e98b284842d12cab6aa23fb.html
19. AI Wiki "Doubao" — https://aiwiki.ai/wiki/doubao
20. Jimeng AI official website — https://jimeng.jianying.com/
21. FindAiverse "Kling AI" entry (verified 2026-08) — https://www.findaiverse.com/tools/kling
22. 8frame "Kling 3.0 Pricing Explained" — https://www.8frame.co/blog/kling-3-pricing-explained
23. Bairimeng AI official website — https://aibrm.com/
24. chooseai.net "Bairimeng" entry — https://www.chooseai.net/tools/1688
25. PixVerse official website — https://pixverse.ai/
26. PixVerse official blog "PixVerse evolves from a creation tool to a production-grade platform" (2026-03-31) — https://pixverse.ai/zh/blog/pixverse-evolves-from-creation-tool-to-production-platform
27. ToolChase "PixVerse Review 2026" — https://toolchase.com/tool/pixverse
28. Hailuo AI official website — https://hailuoai.com/
29. MiniMax official blog "MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities" — https://www.minimax.io/blog/minimax-h3
30. LUMIENAI "MiniMax H3 Tops Video Editing Charts With Open Weights and Lower Prices" — https://lumienai.com/news/minimax-h3-open-weight-video-model-editing-benchmark
31. Southern Metropolis Daily "MiniMax's third-generation video model open-sourced, video editing capability ranked first globally" — https://www.toutiao.com/article/7669739176429912595/
32. China Daily (Finance) "H3 multimodal large model launched today on RunningHub" — https://caijing.chinadaily.com.cn/a/202607/31/WS6a6c655fa310d709c2fc0e89.html
33. Vidu official website — https://www.vidu.cn/
34. Tencent Cloud Developer Community "Vidu Q3 reference-to-video review" — https://cloud.tencent.com/developer/article/2655977
35. Atlas Cloud "ShengShu Models" — https://www.atlascloud.ai/providers/shengshu
36. Comfy official website — https://comfy.org/
37. AI Wiki "ComfyUI" — https://aiwiki.ai/wiki/comfyui
38. GitHub — sanchitwadehra/Effecient-3D-Character-GenAI-Animation-Pipeline — https://github.com/sanchitwadehra/Effecient-3D-Character-Animation
39. Yipin Weike "AI comic drama storyboard design guide" — https://gonglue.epwk.com/322844.html
40. China Business Journal "Shengshu Technology completes RMB 2 billion in financing: how to break through with giants ahead and pursuers behind?" — https://cj.sina.cn/articles/view/1650111241/625ab30902001g6xy