Disclosure: TechSifted has no affiliate relationship with Character.AI, Pika, or Mistral. This is editorial news coverage.

Three stories from the last few weeks. One platform bankruptcy that could end a four-year run as one of the largest AI companion apps. One video model built specifically for creative iteration. One trillion-parameter open-weight release that Mistral spent three additional weeks improving before shipping the weights.

Character.AI Files for Chapter 11

Character.AI filed for Chapter 11 bankruptcy in Delaware on Monday, October 12. CNBC describes it as "one of the largest artificial intelligence companion apps." In its filing, the company said that in its 4-year lifespan it had built "a large and engaged user base," but that the field of AI entertainment "has proven to be costly to develop and grow." Legal exposure compounded the money problem. The filing cited "high costs, coupled with potential litigation related liabilities and regulatory engagement" as the reason the company could no longer generate enough cash to continue.

Disney has agreed to act as the stalking-horse bidder. The deal is a nonexclusive perpetual license to some of Character.AI's intellectual property for $20 million. Other bidders can top that offer. Character.AI expects to close the sale by late November or early December. Disney will also help with the transition, using staff it has hired from Character.AI.

Character.AI said it expects to keep the platform running through the sale process. CEO Colin Adams said the court-supervised sale "provides the Company with the best opportunity to maximize value for all stakeholders." The company added that the platform could be "wound down" if no buyer emerges.

For context: Google licensed Character.AI's large language model technology in 2024 and rehired the company's co-founders, Noam Shazeer and Daniel De Freitas. Kentucky and Pennsylvania have both sued the company over its chatbots. Earlier this year, Character.AI and Google settled a lawsuit brought by the mother of a 14-year-old Florida boy. Character Technologies, the U.S. owner, listed estimated assets and liabilities between $50 million and $100 million in court documents.

One more data point: in September 2025, Disney sent Character.AI a cease-and-desist letter demanding it stop using Disney's copyrighted characters without permission. That is the same Disney now bidding $20 million for an intellectual property license (Source).

Why it matters: Character.AI built a massive user base and still couldn't make the economics work. A large, engaged audience is not a revenue model when your compute costs are high, your legal exposure is growing, and the co-founders were rehired by Google back in 2024. Disney's $20 million is not an acquisition. It is a floor bid for an IP license from a company that sent Character.AI a cease-and-desist letter in September 2025. What is clear: the platform shuts down if no buyer shows up.

Find out more: NBC News: Character.AI files for bankruptcy, will sell some assets to Disney for $20 million

Pika H3 Flex: 12 References, 5 Audio Stems, 4K Output

Pika released H3 Flex on October 29. It is a post-trained version of MiniMax H3, built for creative iteration. The number that leads the announcement: Flex can generate a 15-second clip with synchronized audio at 768p in as little as seven seconds of transformer computation. Pika is upfront about the asterisk: that figure measures DiT computation in a controlled benchmark, not total end-to-end API latency. Reference inputs, higher-resolution finishing, audio-stem separation, hardware, and service overhead all affect completion time.

The architecture choice matters. Pika post-trained MiniMax H3 to run in six denoising steps instead of 50. Whether quality holds across those fewer steps is something leaderboard scores can only partially answer: Pika says Flex ranks first among open-weight models and seventh overall on DesignArena's text-to-video leaderboard at launch, with the caveat that the position reflects the launch snapshot.

What makes this model different from a pure text-to-video system is the reference workflow. Flex accepts up to 12 reference files combining images, video, and audio in one generation. You can give it a character from one image, a setting from another, and motion from a video clip, then tell it what each reference should contribute. Pika lists the kinds of references it means: a sketch, a character sheet, a product photo, a moodboard, a voice sample, or a video of the movement you want. To me that is closer to how professional motion artists actually work than typing a longer prompt.

Audio ships with five separate stems through the Pika API: Dialogue, Score, SFX, Foley, and Ambience. That means the generated audio is separable in post, not locked into a stereo mixdown.

On resolution: Flex generates natively at 480p or 768p. The 1080p, 2K, and 4K upscaler is a Pika-hosted service. Local inference covers text-to-video, first- and last-frame generation, and multimodal reference-to-video at up to 768p. Pika is releasing the generator adapters and inference code, with the weights on Hugging Face. Flex is available now via the Pika API Club, and Pika said it would reach the Pika creation product on November 2, today.

Why it matters: The reference system is the real development here. Most AI video tools let you describe. This one lets you show. Twelve inputs across images, video clips, and audio samples in a single generation is a real change for multi-shot content, and the audio stem separation means you can actually use the output in post rather than treating it as a rough idea. The local weights also matter: you can run generation at up to 768p without touching Pika's hosted stack, which matters for teams with data restrictions. If you are tracking Runway's bet on Solaris as a comparison point, Pika's direction is different. They are going deeper on iterative human control rather than further toward full automation. What AI video creators are saying in 2026 shows how fast that category is moving.

Find out more: Pika's Introducing Pika H3 Flex

Read the rest of this article on TechSifted →

Originally published at techsifted.com.