Runway Aleph and JSON ads | 10 AI filmmaking news stories of the week

This week was wild. Here are some of the most striking recent news stories in AI filmmaking — with articles and videos to back them up. Hold on tight!

1. Runway Aleph: changing reality with simple sentences

Runway's brand-new AI model, Aleph, offers creators a practical and powerful revolution: editing real or AI-generated footage with natural-language instructions. No more frame-by-frame rotoscoping or heavy style transfers. Aleph lets you:

• Add or remove objects
• Change the lighting, time of day, set or weather
• Generate new angles or extra shots
• Change costumes, style or background
• Replace characters using reference images

Despite a few misses on fast movement, it's a giant leap for directors who want to rework their footage in post-production. Available via "Generate Video Gen 4 Chat Mode" for CPP users.

2. Luma also launches a natural-language video editor

Shortly after Aleph, Luma Labs unveiled Modify Video with Instructions. This new tool lets you upload footage and edit it with simple sentences. In early tests, Luma sometimes transformed the video too much or missed subtle details. But it clearly shows that the big players are betting on an entirely text-driven post-production pipeline.

3. Midjourney now lets you set start and end frames

Midjourney is improving its video generation by allowing start and end frames, for more controlled storytelling. This update lets you:

• Create continuous motion with a consistent style
• Add loops
• Extend scenes shot by shot with prompts and keyframes

Transitions still need finesse, but it greatly strengthens creators' control over their videos.

4. Structured JSON prompting takes off after an IKEA ad

An IKEA ad made with Google's Veo 3 popularised JSON prompting, a structured format for ordering scenes precisely for AI models. This format ensures:

• More consistent, reproducible results
• Better control through logic and formatting
• Better alignment with narrative or brand goals

A free JSON prompt converter is available if you want to try the method.

5. Runway teams up with IMAX to show AI films on the big screen

AI filmmaking is making its way into theatres: Runway and IMAX will screen 10 AI-generated films during the AI Film Festival, in ten cinemas across the United States, from 17 to 20 August.

6. Kling Elements update

With version 1.6 of Kling Elements, you can now upload several references (costumes, sets, objects), select specific areas (face, background, costume) and get stylistically consistent renders. Only available in Kling 1.6 (not yet in 2.1) via the official site, not via Freepik or Krea.

7. "Steal" Chrome extension by Higgsfield

The Steal Chrome extension lets you copy, in one click, the lighting, composition or framing of an image on the web and apply it to your AI creations. A handy trick for filmmakers who want to match a visual reference without describing the camera in detail. Bonus: sound effects or music sometimes appear randomly in the renders.

8. Wan 2.2 in ComfyUI: a powerful open-source model

The open-source video model Wan 2.2, available in ComfyUI, offers better aesthetics, improved text rendering, better camera control and better temporal consistency. It is resource-hungry, but LoRA training is already available and it is compatible with the 2.1 tooling.

9. Ideogram's "Character" model for visual continuity

Ideogram can now generate the same character from several angles using a single photo. Pros: photorealistic renders, multi-shot consistency, ideal for recurring characters. Cons: facial expressions and emotions are still imperfect. Available free online and useful for ensuring visual continuity.

10. OpenAI's GPT-5 on the way

Sam Altman announced on Twitter (X) that GPT-5 would arrive "very soon", probably in August. It should combine GPT-4's reasoning, GPT-4o's performance and agent tools like Operator to offer an unprecedented balance of speed, quality and context. It could transform script generation and pre-production — we'll see.