Google Veo 3.1 ships native audio and frame control – 4–8s clips
Executive Summary
Google’s Veo 3.1 finally flips from “imminent” to live, and it’s the control update filmmakers actually want: native audio, first/last‑frame control, reference‑to‑video, and scene extend. You can pick Fast or Quality in Flow right now and generate 4–8s clips with “Beta Audio” turned on. Following yesterday’s endpoints preview we flagged, the rollout is broad enough to matter in real workflows, not just demos.
Day‑0 partners showed up fast. fal exposed text→video, image→video, and first/last‑frame interpolation with native dialogue and tossed in $20 credits for the first 500 signups. Replicate added 3.1 and 3.1 Fast with tighter prompt adherence plus reference images and last‑frame control, while Freepik is running unlimited generations through Sunday for annual Premium+/Pro and Krea discounted Pro/Max 75%. Lovart opened a free trial until Oct 20, ComfyUI shipped API nodes, a Veo 3.1 Fast Gradio app hit Hugging Face, and developers are already seeing Veo endpoints inside the Gemini API.
Early creator tests back up the pitch: references lock identity and environment, dialogue feels more natural, and scene extension avoids the usual crossfade crutches. Some still rate overall fidelity below Sora 2, but the control surface and day‑one availability across hosts look like the real unlock here.
Feature Spotlight
Veo 3.1 everywhere: control, audio, extensions
Veo 3.1 lands across Flow, fal, Replicate, Freepik, Leonardo, Krea, Lovart, and ComfyUI—bringing native audio, first/last‑frame control, references, and extensions into mainstream creator workflows.
Today’s cross‑account story is Google Veo 3.1 going live across creator platforms with native audio, first/last‑frame control, reference‑to‑video, and scene extension. Multiple hosts added access, promos, and APIs for filmmakers.
Jump to Veo 3.1 everywhere: control, audio, extensions topics📑 Table of Contents
🎬 Veo 3.1 everywhere: control, audio, extensions
Today’s cross‑account story is Google Veo 3.1 going live across creator platforms with native audio, first/last‑frame control, reference‑to‑video, and scene extension. Multiple hosts added access, promos, and APIs for filmmakers.
Creators validate Veo 3.1’s references→video and dialogue realism in early tests
Hands‑on runs show Veo 3.1 adhering closely to reference images in both look and environment, with audio that matches the scene’s acoustics; dialogue delivery and gestures feel more natural than prior versions. See a gallery‑talk test with two references and a stand‑up bit. Reference example Stand‑up test
Not everyone is convinced—some call overall quality below Sora 2—but most agree the new controls and tooling (extend, first/last frame) are strong steps forward. Critical take First/last test
Day‑one credits and promos widen access to Veo 3.1 for testing
Several platforms sweetened launch day with credits and discounts: fal’s “veo3.1” code ($20 for first 500), Freepik’s unlimited window for annual plans, and Krea’s 75% off for Pro/Max users. Credit code Unlimited details Krea announcement
Expect a wave of side‑by‑side tests on identity consistency, motion smoothness, and audio sync as a result.
Higgsfield integrates Veo 3.1 with native 1080p, Draw‑to‑Video, Multi‑Shot, and Director Controls
Higgsfield switched on Veo 3.1 with unlimited generations through Monday and layered its own control suite on top—Director Controls, Draw‑to‑Video, Multi‑Shot—and native 1080p with interpolation between keyframes. Higgsfield launch 1080p note
The pitch: beyond baseline Veo, these tools push shot planning and consistency toward production use.
Krea adds Veo 3.1 with image refs, interpolation, improved audio—75% off for Pro/Max
Krea integrated Veo 3.1 with reference images, frame interpolation, and upgraded audio, and paired the launch with a 75% discount for Pro and Max subscribers. Krea announcement
This gives Krea users a lower‑cost way to test multi‑shot continuity and character consistency workflows on day one.
Lovart launches a Veo 3.1 free trial until Oct 20 with unlimited standard gens
Lovart turned on a Veo 3.1 free trial through Oct 20; upgrading to annual Pro/Ultimate by Oct 23 unlocks a month of unlimited Standard Veo 3.1 & Sora 2 plus 10 daily High‑Spec Veo 3.1 and 10 Sora 2 Pro videos. Free trial info
The push targets commercial creators who need both rapid ideation and high‑spec renders in one workflow.
Reference‑to‑video is the breakout control for identity and style this cycle
Across hosts, reference images (characters, logos, styles) are the most‑used control—locking identity while letting action and camera evolve. Replicate and fal both spotlight the workflow, with creators sharing tight adherence examples. Hosting announcement Endpoint overview
Expect deeper integrations from partner apps (e.g., HeyGen identity) to compound this advantage. HeyGen identity
Runware adds Veo 3.1 and Fast on day 0 with R2V precision and first/last frames
Runware enabled Veo 3.1 and Fast with lifelike motion, synced audio, reference‑to‑video precision, and both first‑ and last‑frame control for smoother transitions; try it in their models hub. Runware launch Models page
The integration targets API‑ready production workflows where identity and scene adherence matter.
Veo 3.1 Fast lands on Hugging Face as a Gradio app
A community Gradio app for Veo 3.1 Fast is up on Hugging Face, offering quick Text→Video and Image→Video trials in the browser (mobile link shared for convenience). HF app Gradio space
This is a low‑friction way for creatives to test start frames and motion beats without leaving their browser.
Hedra rolls out Veo 3.1 for photoreal AI video across any imagined scene
Hedra announced Veo 3.1 support, positioning it as the new photoreal standard with the model’s expanded control and fidelity. Hedra announcement
Hedra users can now combine Veo’s reference and framing controls with Hedra’s creative UX for fast, cinematic renders.
Mobile UI captures confirm Veo 3.1 Fast/Quality options in Flow
Additional screenshots from creators show Veo 3.1 listed alongside legacy Veo 2/3 models, with “Beta Audio” tags and credit guidance—useful for teams budgeting runs at speed vs quality. Flow model picker
This mirrors the web rollout and underscores Google’s push to make Veo’s new controls reachable on every surface. Web picker shot
🌦️ Runway’s one‑click VFX apps keep expanding
Runway ships a fresh batch of Apps focused on VFX—weather, backgrounds, time‑of‑day, and relighting—so editors can transform footage with plain language. This follows yesterday’s Apps debut but adds new, concrete tools.
Runway drops weather, background, time‑of‑day and relight Apps
Runway expands its new Apps with a VFX pack—Change Weather, Change Background, Change Time of Day, and Relight Scene—so editors can transform footage with a single prompt Release thread. Following up on Web rollout that introduced Apps, these tools are live on the web with “get started” pages to try now Apps available.
- Change Weather: Make a sunny day overcast or bring torrential rain with one instruction Release thread.
- Change Background: Move subjects into new scenes without rotoscoping or masks App page.
- Change Time of Day: Turn day into night or dial in magic hour from text App page.
- Relight Scene: Adjust mood and lighting direction with promptable relighting App page.
✨ Grok Imagine: animation tricks and styles
Creators lean into Grok for stylized animation: collage‑based start frames, playful utilities, and mood‑driven shots. Threads showcase adherence to references, anime looks, and quick atmospheric prompts.
Collage hack shows Grok’s tight identity adherence vs Veo 3.1
Creators report that with a single collage start frame, Grok Imagine keeps characters and environments locked while Veo 3.1 drifts—especially evident in a cozy Halloween family scene; this extends the one-image pipeline highlighted earlier, following up on 20 Grok videos which proved the approach scales from a single still. See the side‑by‑side and exact prompt for reproducibility in Collage test results and Prompt details.
Grok Imagine nails poetic OVA anime looks
Style studies show Grok delivering lyrical, motion‑rich anime that evokes art‑house moments—like pieces reminiscent of The Piano—while maintaining a cohesive aesthetic across sequences Anime homage.
Grok’s eerie anime excels at unsettling, analog‑horror vibes
Prompting toward discomfort—think “the walls are screaming at you”—Grok Imagine leans into eerie animation that reads like analog horror, giving storytellers a fast path to truly unsettling tone Horror anime clip.
Simple atmosphere prompt lifts Grok shots: burst windows plus a crow
A tiny insert changes everything: adding “windows suddenly burst open from a gust of wind” and a crow to Grok Imagine prompts reliably deepens mood and motion for more cinematic sequences Atmosphere tip.
‘Add a girlfriend’ in Grok fuels upbeat, memeable animations
The playful “add a girlfriend” feature has become a quick way to brighten timelines and seed meme‑ready, feel‑good beats; creators note it even wins over skeptics when used in casual reels Feature note.
🖊️ Higgsfield Sora 2 MAX + Sketch‑to‑Video momentum
Beyond yesterday’s Enhancer news, today creators highlight MAX’s global availability and the Sketch‑to‑Video flow: draw, and it moves—1080p, no timelines or keyframes. Posts emphasize weight/motion/feeling from sketches.
Sora 2 MAX opens globally on Higgsfield—no regions, queues, or codes
Higgsfield’s Sora 2 MAX is now open to everyone with no regional restrictions, waitlists, or access codes, following up on 1080p launch of the Sketch‑to‑Video flow. MAX also touts built‑in deflickering, temporal stabilization, and upscale modules aimed at cinema‑grade output Access note, Model overview, Deflicker claim, Higgsfield page.
Sketch‑to‑Video: draw once, get 1080p motion with no timelines
Creators emphasize Higgsfield’s draw‑to‑motion workflow: sketches become cinematic movement in seconds, with no timelines or keyframing, and render at 1080p Sketch pitch, Napkin to 1080p.
From sketch to scene with sound: MAX outputs motion with synced audio
Beyond visuals, posts highlight that Sora 2 MAX’s Sketch‑to‑Video generates sequences with synchronized audio, aiming for cohesive mood and performance without separate sound passes Audio with motion, Perfect sound claim.
Sketch signals drive animation: weight, motion, and emotion are interpreted
Users report the system “reads” line weight, motion cues, and emotional tone directly from sketches to influence performance and movement choices—reducing prompt micromanagement during blocking Signal claim.
Sketch‑to‑Video adapts framing: 16:9 for cinema, 9:16 for mobile
The flow automatically respects aspect intent—wide 16:9 for cinematic storytelling or 9:16 for mobile reach—so creators can plan deliverables without re‑authoring shots Aspect guidance.
🖼️ Runware Riverflow 1 for one‑shot image edits
Riverflow 1 arrives as a precision image editor that “thinks like a designer,” handling multi‑text changes, targeted details, and native background removal. Launch includes pricing, mini tier, and a #OneShot challenge.
Runware Riverflow 1 launches with designer‑grade one‑shot edits
Runware released Riverflow 1, a precision image editor that “understands what you mean” to deliver one‑shot, production‑ready edits like multi‑text changes, targeted detail tweaks, and native background removal—following up on Riverflow tease which previewed intent‑aware editing. The model was built in‑house with Sourceful and is live now across Runware’s API and Playground Release thread, with example composites showing complex, localized corrections in a single pass Examples set.
Creators can jump in via the Playground for immediate testing and workflow integration Playground links.
Riverflow pricing lands: $0.066/image (Base), $0.05 (Mini), Pro in early access
Runware detailed day‑one pricing for Riverflow 1: Base at $0.066 per image, Mini at $0.05, and a Pro tier in early access, available first in the API and Playground Pricing post. Direct Playground entry points are live for both tiers to test before scaling into workflows Playground base, and Playground mini.
For teams, this puts a precise, intent‑aware editor at sub‑ten‑cent shots, enabling batch product updates and brand fixes without prompt‑surfing.
Riverflow 1 claims top spot on editing arena with one‑shot accuracy
Runware says Riverflow 1 outperforms other image editing models on the Artificial Analysis arena “the majority of the time,” framing it as a one‑shot powerhouse that “thinks like a designer” to hit intent on the first try Arena claim. The launch thread underscores the same strengths—multi‑text changes, targeted detail edits, and native background removal—now backed by public benchmarks sentiment Release thread.
If the results hold across community tests, this could trim multi‑iteration loops for e‑commerce retouching, packaging swaps, and campaign localization.
Runware’s #OneShot challenge: $1,000 prize and $10 credits to try Riverflow
Runware kicked off a #OneShot challenge: generate a single image with Riverflow and post with #Riverflow #OneShot for a chance at $1,000; like and reshare the launch thread then DM to receive $10 in Runware credit to participate Challenge details. This is tailored to stress Riverflow’s headline claim—getting the edit right in one go—while lowering the cost to experiment for first‑time users.
📣 Ad pipelines: multi‑reference, treatments, and boards
Commercial teams share how they plan and style ads: multi‑reference product+pet comps, color palettes, and storyboard assembly—plus a treatment→start‑frame pipeline organized in Figma. Excludes Veo 3.1 feature coverage.
From treatment to start frames: Wander ad pipeline organized in Figma boards
Filmmaker PJ Accetturo shared a production‑ready ad pipeline: begin with a written treatment to lock tone and characters, then generate precise start frames with dedicated DPs and organize selects in Figma for shot‑level oversight (hours per hero frame if needed) Treatment notes, Figma board, Process thread.
The approach shows how treatments translate into consistent boards before any motion work—useful for brand clients demanding continuity across scenes.
LTX Studio lays out a clean ad pipeline: multi‑reference, composition, palette, storyboard
LTX Studio distilled a practical, repeatable workflow for commercial creatives: start with multi‑reference comps to combine product and pet cleanly, define composition (environment, lighting, camera) to shape energy, lock a color palette for consistency, then assemble a storyboard to refine flow across shots Multi‑reference tip, Composition advice, Palette control, Storyboard step. You can try the full pipeline directly in their app LTX Studio.
Collage‑first brand workflow: Grok Imagine locks identity, angles, and a logo end card
Creator Billy Woodward demonstrates a nimble brand pipeline: seed a single collage with characters and environment for identity‑true shots, generate alternate angles on demand, then finish with a custom animated end card built from the logo Collage method, Alternate angles, End card prompt.
This start‑frame→angle‑variants→brand tag flow is a fast path to consistent, shippable social ads without heavy post.
🎨 Stylized stills: Cartographic Couture + MJ v7 recipes
Image‑first creators share reusable style kits and params. A fashion/editorial “Cartographic Couture” prompt lands with ATL examples, alongside Midjourney v7 parameter collages and seasonal spooky looks.
Cartographic Couture prompt pack lands with striking ATL examples
Azed AI dropped a reusable "Cartographic Couture" prompt for fashion/editorial stills—garments formed from flowing topographic maps with contour lines—paired with multiple ATL style examples you can mirror in your own shoots Prompt and examples.
The pack emphasizes muted base + bold accent palettes, soft focus, and wind‑driven motion cues to sell the Vogue x GIS vibe; reposts are already circulating the recipe for wider remixing Repost reach.
Midjourney v7: cohesive looks from a compact param recipe
A new MJ v7 collage shows how a concise setup—--chaos 15, --ar 3:4, --sref <id>, --sw 500, --stylize 500—yields consistent, richly composed frames across subjects Parameter collage, following up on neon recipe that mapped similar settings to wireframe aesthetics.
Creators can slot their own sref to lock style while letting chaos introduce gentle variation; the share is spreading via additional reposts for quick copying into workflows Further share.
Seasonal spooky looks arrive with sharable params and a mini zine
Bri Guy’s 2×2 spooky grid bundles parameter notes (e.g., a pumpkin‑reaper with --chaos 8, --ar 2:3, --raw, --sref 474790598, --sv 4, --sw 800, --stylize 300) and points to a weekly zine for three more recipes Style grid and prompt.
The pack spans painterly creatures to cinematic portraits, giving art directors and poster designers ready‑to‑use seeds for Halloween campaigns; additional mood pieces keep the aesthetic thread alive in the feed Werewolf still.
Topaz Astra upscales sharpen MJ stills for print and social
James Yeung showcased Midjourney pieces run through Topaz’s Astra upscaler, citing noticeable clarity gains for shareable posts and potential print use Astra showcase, with another highlight labeled "Wonders" also attributed to Astra Astra example. The takeaway for stills creators: finish with an ML upscaler to clean noise, tighten micro‑detail, and maintain coherence at higher resolutions without re‑rendering source art.
🔊 Audio, voice and SFX for creators
Sound pipelines see updates: enterprise voice in agents, auto SFX on upload, and instant audio→video assembly. Useful for editors scoring shorts, trailers, and explainers.
Veo 3.1’s native audio and realistic dialogue land across major creator platforms
Veo 3.1’s audio stack (dialogue, music, SFX) is rolling out broadly, giving filmmakers and editors a one‑model path to sound‑on video. fal shipped day‑0 endpoints with "Realistic Dialogue" plus text→video, image→video, and first/last‑frame controls Day 0 launch, with dedicated endpoint links live for both standard and fast variants Endpoint links. Replicate added Veo 3.1 and 3.1 Fast with improved audio generation and tighter prompt adherence Replicate rollout, while Google Flow model pickers now flag 3.1 Fast/Quality as "Beta Audio" for immediate use Model picker. See the model details in Replicate model.
ElevenLabs voices Salesforce Agentforce for conversational customer experiences
ElevenLabs confirmed it is powering the voice of Salesforce’s Agentforce, bringing production‑grade TTS into enterprise agents showcased at Dreamforce. For creatives building interactive explainers and support flows, this slots humanlike voice directly into agent pipelines without extra glue code Agentforce partnership.
fal adds Mirelo SFX v1.5: upload a video, get a synced sound track back
Mirelo SFX v1.5 on fal turns any uploaded video into a version with auto‑generated, time‑aligned sound effects, returning a new soundtrack in one pass (positioned like MMAudio). This is a fast win for shorts, trailers, and reels needing instant, plausible SFX without manual foley SFX model.
Character.ai’s Ovi I2V gets a 25% price cut on Replicate, with synced voice and video
Replicate discounted Ovi (Character.ai’s image/text→video model with native audio) by 25% through Oct 29, making it a cheaper path to 5‑second, 24‑fps clips with synchronized dialogue/SFX in multiple aspect ratios. The model supports text‑only or text+image inputs for quick music promos, bumpers, and social spots Pricing promo, with capabilities and examples in Replicate model page.
📅 Creator contests, stages, and free effects
Plenty for filmmakers and musicians: a $50k+ music video awards program, Halloween effects with prizes, and a Bay Area generative media conference speaker lineup.
OpenArt Music Video Awards go live with $50k+ and 27 prizes
OpenArt opened submissions for its Music Video Awards, offering $50,000+ across 27 awards, a 10/15–11/16 window, and chances to be featured on Times Square billboards—following up on billboards live noted yesterday. Full entry details, rules, and song usage terms are posted, with artist shoutouts for winners Awards announcement, Submissions live. See the official materials in the rulebook and program page rulebook, awards page.
Generative Media Conference reveals speakers: Katzenberg, Blattmann, Mildenhall
The Generative Media Conference (Oct 24, San Francisco) announced a headline trio—Jeffrey Katzenberg (WndrCo), Andreas Blattmann (Black Forest Labs), and Ben Mildenhall (World Labs)—as part of its creative and technical program Speaker lineup. Black Forest Labs also flagged Bay Area appearances next week around PyTorch Conference activities, underscoring a busy creator calendar Week of events.
PolloHalloween contest launches: iPhone 17 grand prize, $10 gift cards for engaged posts
PolloAI’s Halloween contest is live through Nov 3 with one iPhone 17 grand prize; the first 300 posts that reach 30+ combined retweets and replies earn a $10 Amazon gift card, and Halloween effects are free to use this week Contest details. Entry requires using a Halloween effect, tagging @itsPolloAI, adding #PolloHalloween, and submitting via the form How to enter, submission form. Bonus: a new “Halloween Pet Hat” effect is available to try Free effect.
ElevenLabs Halloween music contest offers $2,000 in prizes
ElevenLabs announced a Halloween‑themed music contest with a $2,000 total prize pool, inviting creators to compose and share entries built with its tools Contest post.
Runware’s #OneShot challenge: $1,000 for the best single Riverflow image
Runware opened the #OneShot challenge: generate an image with the new Riverflow model, post with #Riverflow and #OneShot, and their favorite one‑shot wins $1,000. Like and reshare the launch post and DM to receive $10 in Runware credit to participate Challenge info. Riverflow 1, built with Sourceful, focuses on precise, intent‑aware edits and is available in the Runware playground today Model launch.
Wondercraft’s launch video contest: 10 slots compete for $25,000
Wondercraft is calling for ten launch videos to compete for a $25,000 prize pool, inviting creators to submit short launch films crafted with AI tools Contest teaser.
🧰 Creator dev utilities: Replicate, Comfy, cloud
Platform tweaks that smooth production: API sorting for newest models, cheaper multimodal runs, and hardware/cloud access signals for heavier local or cloud work. Excludes Veo 3.1 feature items.
ComfyUI receives DGX Spark hardware, plans benchmark reports
ComfyUI confirmed a special delivery of NVIDIA DGX Spark hardware and says broader benchmarks are on the way hardware update, following up on DGX support that highlighted fast local creation. For studios eyeing on‑prem acceleration, forthcoming numbers will help size local vs cloud trade‑offs for heavy image/video graphs.
Replicate API adds sort-by-created to surface the newest models programmatically
Replicate’s HTTP API now supports sorting models by creation date, making it easier to automatically discover and test the latest releases in pipelines and cron jobs API sorting note. This is a small but practical win for teams wiring continuous evaluation or nightly refresh flows around fast-moving model catalogs.
Comfy Cloud opens more private beta seats via code drop
ComfyUI is handing out additional private beta codes for Comfy Cloud, inviting the community to request access in‑thread beta codes thread. For creators who prefer cloud runs over local nodes, this expands access to managed, shareable workflows without GPU setup overhead.
Ovi I2V on Replicate gets 25% price cut through Oct 29
Character.ai’s Ovi (text/image→video+audio) is 25% cheaper on Replicate until Oct 29, lowering experimentation costs for cross‑modal pieces and short spots pricing update. Details and examples are on the model page, useful for quick trials or batch runs in production backends Replicate model page.
⚖️ Policy shifts: chatbots and adult content gates
Light but relevant policy/news: California passes first‑in‑nation AI companion safeguards, and ChatGPT outlines upcoming personalities and adult‑content gating. Useful context for voice/narrative apps.
California enacts first AI companion safeguards law (SB 243)
California has signed SB 243, the first U.S. law specifically regulating AI companion chatbots, taking effect January 1, 2026. It mandates clear AI disclosure, protects minors from sexual content, and requires crisis‑response protocols for self‑harm scenarios law summary.
- Chatbots must plainly disclose they are AI, not humans.
- Sexual content is barred for minors; age controls are expected.
- Providers must implement protocols for suicidal ideation, including referrals to crisis services.
- Annual impact reports are required, and users gain a private right of action against noncompliant developers.
ChatGPT to add personalities and age‑gated adult content by December
OpenAI signaled a policy shift: ChatGPT will introduce “personalities” in the coming weeks and enable adult content for verified adults in December, following age‑gating and a policy update framed as “treat adult users like adults” policy screenshot.
- Timeline: personalities in weeks; adult content in December.
- Access limited to verified adults after age‑gating; broader content allowances will be accompanied by updated safeguards.
🧪 Small models and multimodal search to watch
A few research/model signals for creative tooling: Anthropic’s fast/cheap Haiku 4.5 shows up across platforms, and Apple’s DeepMMSearch‑R1 targets better web search flows. Also community flags a fake Gemini spec screenshot.
Claude Haiku 4.5 lands on Replicate and Hugging Face at one‑third cost and >2× speed vs Sonnet 4
Anthropic’s small, fast Claude Haiku 4.5 is now live on Replicate and popping up in Hugging Face pickers, marketed as matching Sonnet 4 performance at a fraction of the price and latency Replicate model page, Hugging Face picker.
Early benchmarks and marketing materials cite one‑third the cost and more than twice the speed alongside strong tool use and vision scores, making it attractive for creative assistants, draft generation, and on‑device‑lean workflows Benchmarks chart. For hands‑on use, see the Replicate model card and examples Replicate model card.
Apple details DeepMMSearch‑R1: on‑demand, multi‑turn multimodal web search for MLLMs
Apple researchers propose DeepMMSearch‑R1, a training and tool‑use framework that lets multimodal LLMs plan and run multi‑turn web searches across text and images on demand, then self‑correct using retrieved evidence—useful for grounded creative references and moodboards Paper thread, ArXiv paper.
The system combines supervised fine‑tuning with online RL and introduces a DeepMMSearchVQA dataset to teach when/what to search and how to reason over results; discussion highlights potential impact on agents that fetch style guides, locations, and factual checks during generation Discussion link.
Community debunks “Gemini 3.0 Pro” pricing screenshot as fake
A widely shared “Gemini 3.0 Pro” pricing card was flagged as fake—critics note missing “experimental” labeling that usually accompanies early DeepMind releases and caution against treating it as an official spec Screenshot critique.
Separate app string sightings hint at a “3.0 Pro” upgrade coming, but without validated pricing or cutoffs; creatives should wait for first‑party docs before planning budgets or migrations App strings.