Why “descriptive, not command-based” fixes ignored prompts (and weird motion)

If you’ve ever typed something like “make the camera dolly in” or “don’t change the face,” you’ve already seen the core problem: commands and negatives often don’t translate cleanly into what a video model can observe and animate.

Runway’s own Gen-4 guidance pushes creators toward prompts that describe what happens on-screen, with special emphasis on motion. The Gen-4 Video Prompting Guide explicitly recommends using the text prompt to focus on describing motion, and it also recommends starting simple and iterating by adding details only as needed. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

The same guide also recommends positive phrasing and warns that negative phrasing is not supported and may create unpredictable results. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

This matters for Veo3Gen users because the habit that Runway is trying to train—describe observable change over time—tends to produce cleaner motion and fewer “ignored” instructions across image-to-video workflows.

One more practical detail from Runway: Gen-4 generates video in 5- or 10-second durations based on an input image plus a text prompt. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide) That timebox is why wording that implies pace (slow vs sudden) and what changes first often beats a pile of commands.

The 3-part rewrite method: Subject → observable action → camera behavior

When you catch yourself writing commands, translate them into a three-part “shot description” that a model can animate:

Subject (who/what): keep it general when possible (e.g., “the subject”), which Runway recommends. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)
Observable action (what changes): movement, expression shift, interaction, deformation, particles, lighting change.
Camera behavior (how it’s seen): framing change, parallax, focus shift, stability, lens feel—described as visible outcomes, not operator instructions.

If you’re working from an input image, remember the prompt’s job is usually to explain motion, not re-list what’s already visible. ImagineArt’s Runway Gen-4.5 guide phrases this directly: the prompt’s main job is to describe motion rather than repeating what is already visible in the image. (https://www.imagine.art/blogs/runway-gen-4-5-prompt-guide)

Also: start simple, then add one new element at a time—Runway suggests this as a troubleshooting method to identify what helps or hurts. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

21 command-style prompts rewritten into descriptive shot language

Below is a steal-this library. Column B rewrites commands into descriptive phrasing aligned with Runway’s guidance (motion-forward, positive, simple). Column C adapts the same idea into a Veo3Gen-friendly variant—still descriptive, just a bit more explicit about what must remain consistent.

A) Command-style line	B) Descriptive rewrite (Runway-style)	C) Veo3Gen-friendly variant
1) “Dolly in on the subject.”	“The subject grows larger in frame as the background subtly expands; gentle forward movement.”	“Slow push-in: the subject increases from medium shot to close-up; stable framing.”
2) “Pan left to reveal the city.”	“The frame drifts left, gradually revealing more of the city skyline on the right edge.”	“Smooth leftward drift; skyline enters frame progressively; no sudden jumps.”
3) “Add handheld shake.”	“Slight handheld micro-jitter with natural sway; imperfections feel human.”	“Subtle handheld sway, mild shake; keep subject readable.”
4) “No shaky camera.”	“Locked-off tripod shot; stable frame with zero jitter.”	“Static camera, stabilized footage feel; consistent framing.”
5) “Rack focus to the background.”	“Focus shifts from the subject to the background; foreground softens while distant details sharpen.”	“Noticeable focus pull: subject blurs, background becomes crisp.”
6) “Zoom in quickly.”	“Rapid framing change: the subject abruptly fills the frame within a moment.”	“Fast push-in/zoom feel; end on tight close-up, stable.”
7) “Make them smile.”	“The subject’s neutral expression transitions into a warm smile.”	“Expression change only: neutral → smile; keep identity consistent.”
8) “Make them look surprised.”	“Eyebrows lift and eyes widen; the subject reacts with visible surprise.”	“Surprised reaction with widened eyes; no wardrobe change.”
9) “Change outfit to a red jacket.”	“The subject is wearing a red jacket throughout the shot; fabric folds move naturally.”	“Wardrobe stays a red jacket for the full clip; consistent colors and logo placement (if present).”
10) “Don’t change the face.”	“The subject’s facial features remain consistent across frames; only expression and head movement change.”	“Identity consistency: same face and hair; avoid morphing; only natural motion.”
11) “Make the hair blow in the wind.”	“A light breeze lifts and sways the subject’s hair; strands flutter intermittently.”	“Wind-driven hair motion; consistent hairstyle and color.”
12) “Add rain.”	“Rain falls diagonally through the frame; droplets streak past the lens; surfaces darken slightly.”	“Steady rain with visible streaks; keep lighting consistent, no sudden scene change.”
13) “Make fog roll in.”	“Low fog creeps across the ground, gradually thickening and softening distant contrast.”	“Fog density increases over time; background fades slightly, subject remains clear.”
14) “Add sparks.”	“Small bursts of sparks flicker near the source, scattering outward and fading quickly.”	“Short spark bursts; sparks arc then extinguish; keep overall exposure stable.”
15) “Make it glow.”	“A soft luminous glow intensifies around the object, blooming gently without washing out details.”	“Glow ramps up subtly; preserve texture; no blown highlights.”
16) “Explode the wall.”	“The wall fractures, debris ejects outward, dust billows, and fragments tumble to the floor.”	“Single controlled blast: debris and dust expand then settle; keep camera stable.”
17) “Show product rotation.”	“The product rotates slowly on its vertical axis; highlights glide across its surface.”	“Smooth 360-style partial turn; consistent label alignment and readable silhouette.”
18) “Make the background move faster.”	“Foreground stays steady while background motion streaks increase, suggesting higher speed.”	“Speed-up feel: background blur/parallax increases; subject remains centered.”
19) “Make it cinematic.”	“Soft contrast, controlled highlights, gentle depth-of-field; calm, intentional motion.”	“Cinematic tone: shallow DOF, smooth motion; avoid chaotic movement.”
20) “Avoid motion blur.”	“Crisp edges during movement; clear detail with minimal blur.”	“Sharp motion: reduce smear; stable shutter-like clarity.”
21) “Add camera orbit around subject.”	“Perspective shifts in an arc; foreground parallax increases as the camera circles the subject.”	“Slow orbit: background slides laterally; subject remains central; smooth parallax.”

Camera-move implication examples you can reuse (quick list):

“The subject grows larger in frame…” (push-in)
“Foreground parallax increases…” (orbit / lateral move)
“The frame drifts left/right…” (pan-like reveal)
“Perspective shifts upward as the horizon drops…” (crane up feel)
“The subject moves from center to edge as framing re-centers…” (reframe/track)

Motion-first prompting: what to describe when you only have 1–2 lines

Runway’s Gen-4 Video Prompting Guide explicitly says to use the text prompt to focus on describing motion. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

So if you only get one or two lines, prioritize:

What moves first: “A breeze starts, then hair follows.”
Direction + pace: “slow drift,” “sudden jolt,” “gradual rise.”
What remains constant: identity, wardrobe, framing stability, lighting mood.

And because Runway recommends starting with a simple prompt and iterating, treat your first generation like a baseline, not a final. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

Positive phrasing swaps (what to say instead of “no / don’t / avoid”)

Runway warns that negative phrasing is not supported and may produce unpredictable results—so swap negatives for a positive, visible target. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

Here are quick conversions you can paste:

“Don’t make it shaky” → “Locked-off tripod shot; stable frame.”
“No motion blur” → “Crisp edges during movement; minimal blur.”
“Don’t change the face” → “Facial features remain consistent; only expression changes.”
“No outfit changes” → “Same outfit throughout; consistent colors and materials.”
“Avoid weird warping” → “Natural anatomy and proportion; subtle, realistic motion.”

Mini checklists: 5-second clips vs 10-second clips (how your wording changes)

Runway Gen-4 outputs 5- or 10-second durations. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide) Even if you’re generating elsewhere, this is a useful mental model: short clips like one beat; longer clips can support a mini-arc.

For ~5 seconds

Describe one primary motion (push-in or expression change).
Use a clear end-state: “ends in a close-up.”
Keep secondary effects minimal (one particle system, not three).

For ~10 seconds

Add a second phase: “fog rolls in, then settles.”
Describe timing words: “gradually,” “halfway through,” “at the end.”
Confirm what stays constant (identity, wardrobe, environment).

Copy-paste prompt blocks you can adapt inside Veo3Gen today

Use these as starting points, then iterate by changing one element at a time (Runway recommends adding one new element at a time for troubleshooting). (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

Block 1: Clean push-in with expression change

“Stable, locked-off feel with a gentle forward move. The subject grows larger in frame from medium shot to close-up. The subject’s expression shifts from neutral to a small smile; facial features remain consistent throughout.”

Block 2: Weather + environment motion

“A light breeze moves through the scene; leaves and small details sway naturally. Low fog creeps along the ground and gradually thickens, softening distant contrast while the subject remains clear.”

Block 3: Product demo rotation

“The product rotates slowly on its vertical axis; highlights glide across the surface. Background remains clean and consistent; crisp edges and stable framing.”

5-second self-audit checklist (before you hit Generate)

Observable in-frame: Can every key instruction be seen (not inferred)?
Change over time: Did you specify what moves, in what direction, and how fast?
Camera behavior: Did you describe framing/parallax/focus outcomes rather than “do X”?
Positive phrasing: Did you replace “don’t/no/avoid” with a visible target?
Constants: Did you state what must stay consistent (identity, outfit, lighting, background)?

FAQ

What does “focus on describing motion” actually mean?

It means your text should explain what changes from frame to frame (subject motion, camera motion, scene motion), because Runway’s Gen-4 Video Prompting Guide recommends using the prompt to focus on describing motion. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

Should I use negative prompts like “no blur” or “don’t change the face”?

Runway’s guide recommends positive phrasing and says negative phrasing is not supported and may produce unpredictable results—so rewrite negatives into visible positives (e.g., “locked-off tripod shot; stable frame”). (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

How detailed should my prompt be?

Runway recommends starting with a simple prompt and iterating, adding more details only as needed. (https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide)

Should I repeat everything I see in the input image?

In image-to-video workflows, it’s often more effective to describe the motion you want rather than re-stating the static image; ImagineArt’s Gen-4.5 guide frames the prompt’s main job as describing motion rather than repeating what’s already visible. (https://www.imagine.art/blogs/runway-gen-4-5-prompt-guide)

Build this into your Veo3Gen workflow (CTA)

If you want to automate these rewrites (command → descriptive → variant) inside your pipeline, you can integrate Veo3Gen directly via the API and standardize prompt style across your team.

For teams generating lots of iterations, compare tiers on the pricing page and pick a plan that matches your render volume and experimentation cadence.

Runway Gen‑4’s Official “Descriptive, Not Command‑Based” Rule — 21 Creator Rewrites You Can Steal for Veo3Gen (as of 2026‑05‑03)