Seedance Omni References.

PROMPT 1: Change the eye to look like a snake eye

PROMPT 2: Change the weather to a thunderstorm with heavy rain

PROMPT 3: Change the time period to the 1920's

PROMPT 4: replace the cow in
reference
with the bear in
reference


PROMPT 5: As the woman walks, a massive alien spacecraft falls from the sky and crashes into the ground. The spacecraft is way in the distance, but it's massive so we see if crashing into the earth.

PROMPT 6: replace the asteroids with meatballs covered in marinara sauce. Replace the smoke and fire with marinara sauce and spaghetti noodles. Rather than dirt and debris, have the impacts produce marinara sauce and spaghetti noodles.

PROMPT 7: A massive creature emerges from the clouds in the distance. The creature is so massive that it passes through the clouds.

Kling Motion Control was introduced in 3.0.
1080p and 30 seconds (!)
Try to tell which one here is generated :)

New LTX 2.3
You can try it here: https://app.ltx.studio/ltx-2-playground/t2v
Interestingly, the parameters include duration up to 20(!) seconds, 4K resolution, and even 50 fps. Camera motion is also separated into its own setting parameter with lots of options.
The weights have already been released: https://huggingface.co/Lightricks/LTX-2.3
Support has also been added to Comfy workflow templates: https://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_ltx2_3_t2v.json

Animation generation in After Effects using GPT-5.4.
The Atom plugin is used here. https://tryatom.ai/
Under the hood, it runs MCP connected to After Effects.
After Effects also has a version with Gemini and lots of other fun stuff.

ByteDance, the parent company of TikTok, has postponed the global launch of Seedance 2.0. The reason is copyright claims from major Hollywood studios and streaming platforms. Reuters, citing The Information, reports that the launch—originally planned for the end of March—has been put on hold, while the company has not yet publicly provided a detailed confirmation of these reports.

Midjourney V8
The V8 alpha has been released. In short, it’s undertrained, still struggles with fingers, there’s no editing yet, and the video model has been postponed (according to the developers).
I was going to write about the new features and how Midjourney is falling behind, but I found a great tweet that sums it up perfectly and matches my thoughts.
🎨 Midjourney launched the V8 alpha on March 17, 2026. http://alpha.midjourney.com is positioned as the fastest model yet, with better prompt understanding, improved text rendering, support for moodboards, sref, multiple aspect ratios, native HD 2K output, and image generation about 4–5x faster than previous versions.
🤖 The problem is that, based on my tests, these promises don’t hold up. Incorrect hands, distorted proportions, weak anatomy, and overall quality that often feels worse than V7. What stands out most is that in a rapidly evolving AI image market, releasing such a raw alpha risks looking more like a response to competition than real progress.